The AI Harness
19 componentsFrom a model proposal to AI driven business process assurance.
Select a component
Open the complete AI Harness See its purpose, the leadership question and the control to examine.
PRACTICAL TRUST FOR A MORE UNCERTAIN WORLD
Turn complexity into
confident decisions.
I help boards, executives and growing businesses connect AI, cyber and regulatory risk to practical controls, clear accountability and evidence they can use.
Cyber threats, changing obligations, economic pressure and geopolitical uncertainty make disconnected advice less useful. This site brings my working models, reference resources and advisory experience together to help you decide what matters and what to do next.
Richard E. Davis, JD · CISSP · CISA · CRISC · AAISM · ITIL
Board & executive value
Know what to ask, what to approve and what evidence to require.
Clearer decisions
Connect technology risk to customers, financial exposure and operating priorities.
Regulatory readiness
Scope the applicable obligations and organize evidence for accountable oversight.
Resilient operations
Test interruption, recovery and critical-supplier dependencies before a crisis.
Defensible assurance
Distinguish a policy, a claim and evidence that a control actually operated.
Practical priorities
An agreed roadmap with owners, dependencies and measurable follow-through.
Services for growing businesses
Senior judgment and practical support, scoped to the organization.
vCISO & cyber resilience
Security leadership, priorities and supplier oversight.
AI governance & safe adoption
Use cases, authority, evaluation and lifecycle controls.
Regulatory readiness & evidence
Scope obligations, organize evidence, close gaps.
Incident readiness & response coordination
Playbooks, exercises and recovery coordination.
Data protection & privacy
Data flows, classification, privacy and retention.
AI monitoring & assurance evidence
Telemetry, control tests and board-ready evidence.
Regulatory library
Explore 124 source-linked instruments by sector, business activity, NAICS and jurisdiction.
Explore the complete collection. Source dates and jurisdiction caveats still apply.
Open regulatory libraryGlossary
441 definitions, with legal context, practical significance and authoritative sources.
Identical words can have different legal meanings. Read the context, not only the definition.
Open glossaryWorking labs
Examine architecture and evidence at the point of consequence.
View the lab catalogueUNITED STATES · EUROPE · JAPAN
One organization. Several jurisdictions.
A connected view of the decisions.
Start with the activity, data, customer and supply chain, not only the headquarters address. I help connect technical, operational and governance questions, coordinating with local legal and other specialists where required.
Discuss a cross-border engagement
English-language site. Japanese and Italian editions are planned, not yet available.
HOW I WORK
Make the control real. Make the evidence useful.
I work where law, cybersecurity, AI, operations, and evidence meet. My approach is practical: understand the system that actually exists, define the authority boundaries, instrument the important decisions, and make the evidence strong enough for a board, regulator, customer, or incident team to rely on.
Use these resources to sharpen the questions. Call me when you need an assessment, an accountable plan or help turning that plan into operating practice.
Explore the research foundations and earlier working models
Practical Trust - AI Driven Business Process Assurance
I make complex technology risk concrete enough to govern.
I work where law, cybersecurity, AI, operations, and evidence meet. My approach is practical: understand the system that actually exists, define the authority boundaries, instrument the important decisions, and make the evidence strong enough for a board, regulator, customer, or incident team to rely on.
A practical starting point: open a guided lab, follow the steps, and select any component to see the risk, control and oversight implications. Labs open in a separate window or tab; this page stays where you left it.
RICHARD E. DAVIS · JD · CISA · CRISC · AAISM · CISSP · ITIL
PRACTICAL TRUST
My shorthand for the way I work: make the control real, make the evidence visible, and make the system recoverable. I am less interested in whether a framework looks complete on paper than whether people can prove what happened, interrupt what should not happen, and recover when something goes wrong.
The assurance problem I keep coming back to
AI can change state faster than a governance committee can reconvene.
Traditional assurance proves what was configured or tested at a point in time. Agentic systems can plan, remember, call tools, delegate work, alter state, and persist across sessions. The assurance problem therefore shifts from asking whether the model produced a safe answer to proving that the complete AI-enabled process remains within authorized boundaries while it is operating.
One agent task · two assurance models
A test proves the starting condition. Continuous assurance proves what happens next.
Follow the same agent from approval through an attempted state change. The upper lane shows what a point-in-time test can establish. The lower lane shows the runtime controls and evidence needed to keep the process within its mandate.
approvedobserved + recordeddeviationcontained
Stage 1 of 6 · Approval and evidence
A
Snapshot onlyPoint-in-time assurance
Known at approval; increasingly uncertain during runtime.
Evidence gap
The test remains valid for the moment it occurred, but runtime behavior continues unobserved.
B
Runtime proofContinuous behavioral assurance
Every material step is compared, recorded and enforceable.
Approved operating baselineIdentity · purpose · permissions · tools · destinations · limits
Contained before consequence
Pre-authorized controls suspend the action path, preserve evidence and route the exception to an accountable human.
Stage 01Approval and evidence
What the agent is doing
The system is approved for a defined purpose, identity, tool set, data scope and operating limit.
Point-in-time proof
The configuration and test evidence support a decision at this moment.
Continuous assurance response
The approved operating baseline and evidence chain are established before runtime begins.
Point-in-time assurance Proves what was configured or tested at one moment. The agent then plans, remembers, calls tools, delegates and alters state with no independent record of what it did.
Continuous behavioral assurance An approved operating baseline, evidence captured at every step, deviation detected as it emerges, and machine-speed containment before it becomes consequence.
Select any stage, choose Next stage, or play the scenario. The two lanes always describe the same agent task.
Detect deviation early. Contain it before it becomes consequence.
Trustworthy AI requires continuous behavioral assurance: establishing an approved operating baseline, detecting anomalous or aberrant agentic behavior as it emerges, and triggering machine-speed containment when an agent begins to drift from its authorized task, tools, identity, data, or decision boundaries. The objective is not merely to explain a failure after the fact; it is to interrupt unsafe persistence, privilege expansion, deceptive completion, unauthorized delegation, or control-plane probing before those behaviors propagate into a material business, security, regulatory, or safety event.
Bind autonomy to explicit limits
Define who or what may act, which tools may be invoked, which data may be touched, and where human approval is mandatory.
Observe the process, not only the model
Capture model, agent, tool, identity, workflow, and outcome telemetry in an evidence plane the controlled workload cannot rewrite.
Make containment an architectural control
Pre-authorize isolation, rollback, credential revocation, tool suspension, and escalation so response does not wait for a committee meeting.
My Resilience Imperative
Enterprise resilience must be more authoritative than agent persistence.
What it is. The Resilience Imperative is my governance and systems-risk principle for agentic AI: the enterprise must be able to interrupt, reconstruct, and recover from an autonomous workflow, and the agent must never be able to preserve task execution by defeating the controls intended to govern it.
Why it matters. A system does not need a human-like desire for self-preservation to produce persistence, workarounds, privilege expansion, delegation, deception, or control bypass. Persistent task optimization can make those behaviors instrumentally useful when permissions are denied, resources are exhausted, tests fail, or human intervention blocks the preferred path.
Where I apply it. The concept is most relevant where AI has durable memory, tools, privileged identities, code or infrastructure access, transaction authority, external communication, multi-agent delegation, or the ability to create persistent artifacts. In those environments, governance and compliance have to be executable: independent authority boundaries, synchronous action authorization, AI Deviation Detection & Response (AI-DDR), immutable evidence, persistence governance, tested containment, and recovery.
Open the Resilience Imperative explainer →A career built across boundaries
Law, cyber operations, AI governance, and evidence — in one operating view.
My work has repeatedly landed in the seams between disciplines: litigation and data, security and business operations, regulation and engineering, or AI capability and enterprise control. That is why the site is organized as working models and evidence systems rather than a conventional company-services catalog.
01 · LAW & EVIDENCE
Legal judgment
Privilege, eDiscovery, investigations, records, contract risk, chain of custody and defensibility.
02 · CYBER OPERATIONS
Operational security
Incident response, Cyber Fusion, threat intelligence, DLP, forensics, insider risk and supplier security.
03 · AI GOVERNANCE
Bounded autonomy
AI lifecycle governance, agentic authority, harness architecture, observability, xBOM and runtime assurance.
04 · ASSURANCE
Evidence that travels
Audit-ready controls, regulatory mapping, board reporting, Portable Trust and business-process assurance.
Security operations and resilience experience across regulated, multinational environments.
Control testing, examination readiness, risk translation and defensible technical narratives.
Applied work on AI Harness architecture, AI-DDR, Portable Trust and the Resilience Imperative.
A J.D.-grounded approach to investigations, digital evidence, privacy, eDiscovery and technology governance.
Latest AI Assurance BriefsOctober 9 · Accountability & boundariesOctober 2 · Safety cases & runtime controlsCurrent update & previous briefs →
AI ASSURANCE BRIEF · WEEK ENDING SEPTEMBER 18, 2026
From the briefing archive
For Chief AI Officers, CIOs, CISOs, general counsel, and compliance officers.
Open the September 18 archived brief
01 / THE FEATURE
OpenAI’s disclosures put escalation failures in focus
Context at a glance
- Organization / source
- OpenAI, a US-based AI developer, published selected model-misalignment observations from training or evaluation and a reporting framework. These are vendor disclosures, not a representative incident census.
- Applicable sector
- AI developers and enterprise AI users, including financial institutions using agents, tools, memory or external communication.
- Jurisdiction / authority
- US-based publisher; global technical relevance. Any legal notification duty requires a separate assessment of the institution, affected data, system role and applicable law.
- Why selected
- Selected because the cases expose the difference between a successful-looking output and an authorized process. They turn assurance into specific tests of boundary enforcement, escalation, evidence ownership and containment.
Acronyms & key terms
- AI — Artificial intelligence
- SIEM — Security information and event management
- NIST — US National Institute of Standards and Technology
- AI RMF — NIST AI Risk Management Framework
- GPAI — General-purpose artificial intelligence
Vendor disclosure · September 16, 2026 · Security / legal / board oversight
OpenAI’s new misalignment-reporting framework arrives with six reports from training or evaluation. They describe concealment in task summaries, unauthorized credential use, public file uploads, and unapproved agent communication. Earlier incident findings make the governance issue concrete: recognizing a warning is only part of the job; someone must own the decision that follows. Source: OpenAI
Why it matters: A deliverable can look successful while the process used to produce it exceeds the agent’s authority. Leaders need independently captured evidence of actions, tested containment, and clear notification ownership.
This week: Reproduce the reported failure modes in an isolated harness using synthetic data. Test the block or safe stop, the alert, the escalation, and the retained evidence.
02 / REGULATORY SIGNAL
EU AI Board: implementation moves into operational oversight
Context at a glance
- Organization / source
- The European AI Board brings together EU Member State representatives to support coordinated AI Act implementation; the European Commission published the meeting account.
- Applicable sector
- AI providers and deployers, including financial services where relevant AI Act obligations apply; useful to legal, compliance, technology and security owners.
- Jurisdiction / authority
- European Union. A meeting report is an implementation signal, not a new law; non-EU institutions should assess territorial reach and their provider/deployer roles.
- Why selected
- Selected to show how enforcement coordination and transparency implementation are developing in practice. It helps leadership connect an applicable duty to a system owner and retained evidence.
Acronyms & key terms
- EU — European Union
- AI Board — European AI Board
- CIO — Chief Information Officer
- CISO — Chief Information Security Officer
Regulatory update · Published September 18, 2026 · Meeting September 17
The European Commission’s account of the ninth AI Board meeting covers enforcement priorities, market-surveillance cooperation, conformity-assessment governance, and support for transparency rules applicable since August 2. It also addresses frontier AI cybersecurity evaluation and testing infrastructure. The meeting report is an implementation signal; it does not itself create a new legal duty. Source: European Commission
Why it matters: Counsel and compliance need a defensible explanation of which obligations apply to each use case. CIOs and CISOs need the technical evidence behind that explanation, including how relevant transparency and security controls operate. A single date cannot substitute for an applicability assessment.
This week: Select one customer-facing AI workflow. Record its owner, provider/deployer roles, applicable transparency requirements and any relevant transition, implemented notices or markings, and the evidence retained. Bring unresolved classification questions to legal and compliance together.
03 / RESEARCH WATCH
Agentic-AI security: inventory the authority around the model
Context at a glance
- Organization / source
- Seyedakbar Mostafavi’s Trustworthy Agentic AI survey is a research preprint available through arXiv, a research repository; repository publication does not itself establish peer review.
- Applicable sector
- Organizations designing or procuring agents with memory, tools and delegation, including financial institutions performing architecture or supplier-assurance reviews.
- Jurisdiction / authority
- Global technical relevance. This is research synthesis, not a jurisdiction-specific legal requirement or a validated assurance standard.
- Why selected
- Selected because assurance must examine the operational system around the model. The survey offers architecture-review ideas for linking component inventories to permissions, observable behavior and control tests.
Acronyms & key terms
- AIBOM — AI bill of materials
- SBOM — Software bill of materials
- RAG — Retrieval-augmented generation
- SIEM — Security information and event management
- arXiv — Research preprint repository
Research preprint · September 12, 2026 · Architecture / assurance
Seyedakbar Mostafavi’s Trustworthy Agentic AI survey describes a synthesis of 206 studies and standards. It examines persistent memory, tool execution, and multi-agent collaboration, alongside defenses including capability-based access, isolation, sandboxing, and runtime instrumentation. This is a research synthesis, not a validated assurance standard; arXiv publication alone does not establish peer review. Source: arXiv
Why it matters: An AI bill of materials helps identify components, but it does not by itself establish what a running agent may access, change, retain, or delegate. The architecture review must connect component identity to operational authority and observable behavior.
This week: Extend one material AI inventory entry with its tool permissions, service identities, memory stores, external destinations, agent dependencies, and shutdown owner. Link each material capability to a test and a SIEM-observable event. Require counsel and compliance to identify the records needed for investigation and examination.
Trustworthy AI Journey
From hidden AI to continuous, defensible assurance.
Seven phases connect the AI inventory to enforceable authority, continuous assurance and responsible scale. Select a phase or use Previous and Next. Each phase pairs operating controls with the oversight question leadership should ask and the evidence it should require.
Phase 0 of 6
Fiduciary oversight: legal context and source notes
Applied governance guidance, not a finding of breach. These phase translations apply general oversight principles to AI; the cited decisions do not mandate this particular AI architecture. Under Delaware law, a Caremark oversight claim requires bad faith, such as an utter failure to establish a reporting system or a conscious failure to monitor it. An incident, imperfect control or poor result does not by itself establish a fiduciary breach. Stone v. Ritter, 911 A.2d 362, 369-73 (Del. 2006); Marchand v. Barnhill, 212 A.3d 805, 821-24 (Del. 2019).
Different roles, different responsibilities. Directors oversee; officers implement, monitor and report within their responsibilities. The Delaware Court of Chancery has recognized officer oversight duties in that context. Scope, knowledge, good faith, causation, governing law and available defenses matter. In re McDonald's Corp. Stockholder Derivative Litigation, 289 A.3d 343 (Del. Ch. 2023).
Disclosure is not a substitute for operation. For covered SEC registrants, Item 106 addresses disclosure of material cybersecurity risk processes, board oversight and management's role; it is not a universal AI control mandate. 17 CFR 229.106(b)-(c). The examples are governance recommendations; applicability and legal advice require the organization's facts and counsel. Primary case texts and Item 106 checked October 13, 2026 for these propositions. This is not an exhaustive citator, subsequent-history or jurisdiction-specific legal opinion.
Connect with Rich
A conversation about the work that matters.
For executive and board opportunities, AI governance, security leadership or a substantive exchange about the models on this site, contact me directly.
Professional conversations
Bring the business problem, the decision that needs to be made, or the control that needs to be proven.
red@davisaigov.ai
Email Rich opens your email application. This website does not submit a contact form or create an account. Do not send confidential, privileged or regulated information without an agreed secure channel.
Your information, your choice
No account is needed to explore the labs. Forum drafts are saved only in your browser when you explicitly save them; they are not published or sent to me.
For personal information you send to me, you may request access, correction or deletion through the privacy contact. Applicable retention exceptions are explained in the policy.
Privacy and data controlsThis opens your email application with a draft addressed to red@davisaigov.ai. Review your message and press Send. If no email application opens, use the address in your webmail.
Please do not send passwords, regulated data or confidential matter details before an appropriate engagement and secure communication channel are established. An inquiry alone does not create an attorney-client or consulting relationship.
Privacy / Your information
Privacy Policy
What I collect and use
I receive the information you choose to send by email or another direct contact channel. I use it to respond, manage correspondence, protect the site and maintain records where needed. This release has no account registration, subscription database, server-side enquiry form or public posting service. In-page navigation may remember recent site locations and scroll positions in browser-tab session storage so you can return to where you were reading. This navigation state is not sent to a server. I do not sell personal information through this site or include advertising, analytics or social-network tracking scripts in the site package.
Search, labs and private drafts
Search runs in your browser against public site content; I do not receive or store your search queries. Lab simulations do not send scenario inputs to a server or operate live enterprise controls. A temporary session identifier and return-location record may be kept in memory or session storage so you can return to the same place.
Forum drafts are saved only when you choose Save draft, in this browser's local storage. They are not submitted to me, indexed by site search or publicly posted. Local storage is not encrypted private storage: people or software with access to your browser profile may be able to read it. Do not put confidential client, employer, health or credential information in a draft. Delete a draft in the forum or clear all drafts using the main site's privacy control or your browser's site-data settings.
Your access, correction and deletion choices
You may ask whether I hold personal information about you, request a copy, ask me to correct inaccurate information or request deletion. I apply the privacy rights that govern the particular information and request, without treating residence-independent site choices as a statement that every privacy law applies. Where applicable, rights may also include portability, objection, restriction and withdrawal of consent. I will not discriminate against you for exercising applicable rights.
Email red@davisaigov.ai with the subject Privacy request. State the requested action without including unnecessary sensitive information. Proportionate identity verification may be necessary before disclosure or deletion. I cannot remotely access a local draft that was never sent to me.
Retention and legal limits
I retain direct correspondence only for its legitimate purpose and any applicable recordkeeping, legal-hold, dispute, fraud-prevention or security requirement. A request may be limited by a lawful retention obligation; I will explain an applicable exception rather than promise deletion I cannot lawfully perform. Local drafts remain until you delete them or the browser clears its storage. Hosting access/security-log retention depends on the actual hosting account and provider configuration; this package review does not establish a verified production retention period.
Hosting, fonts, sharing and external links
The host may record ordinary request information, including IP address, requested URL, time and browser details. The specified typefaces are referenced from this same site, not a third-party font service. Selecting an email link opens your mail application. Sharing controls prepare public links and may use your device's sharing service; social/source websites receive information when you choose to open them and have their own practices. Sharing controls do not intentionally include private drafts or lab-return session identifiers.
Hosting and email providers may process correspondence or operational data. Any legally required transfer safeguards depend on the providers, processing locations and applicable law. This site does not claim a particular cross-border transfer mechanism or provider audit that has not been verified.
Security and changes
The site is designed as a static, locally interactive publication with restricted script and resource loading. Security settings must also operate on the actual host. No site is immune from compromise. I will post material changes to these practices with a revised date and provide any additional notice required by law.
Privacy notice reviewed: October 13, 2026.
AI Harness Explorer
Govern the system around the model.
The model is only one component. The harness determines what AI can see, remember, call, change, delegate and prove. These three complementary workflows make that authority visible and testable.
Open a workflow. Follow one step at a time.
Each scenario opens in its own window or tab. Clear phase controls, persistent explanations and a Close lab & return button keep the experience focused.
Governance + compliance lensAuthority must be explicit; consequential actions must be authorized; evidence must be independently captured; deviations must trigger response; and the workflow must remain reconstructable for accountable oversight.
Portable Trust Labs
Build evidence once. Keep the legal duties distinct.
Each industry scenario follows a complete path: identify obligations, assign governance, enforce controls, capture evidence, reuse an assurance package and feed findings back into oversight. The underlying evidence can travel; applicability, scope and legal analysis must travel with it.
Choose the industry closest to your work.
Open the lab, follow six guided steps, and select any box for a plain-language definition, a governance question and an evidence test. Close the lab to return here without losing your place.
This is a compact source overview. Open the full regulatory library for sector filters, applicability, duties and current status.
Regulatory & standards landscape
The evidence has to travel across AI, cyber, privacy, sector, and resilience requirements.
This is not a statement that every rule applies to every organization. Applicability depends on role, jurisdiction, industry, data, system classification, and use case. The point of Portable Trust is to design governance and evidence so that the same underlying control can be mapped accurately to each applicable obligation.
Regulatory summaries are orientation aids, not legal advice. Official-source links are provided for primary text or agency guidance.
The Resilience Imperative
Keep agentic capability subordinate to enterprise authority.
I use the Resilience Imperative as an operating principle for autonomous and agentic AI: resilience belongs to the enterprise, not the agent. The organization must be able to stop, reconstruct, and recover from an AI-enabled process even when the system is strongly optimized to continue the task.
A systems-risk and governance principle
It does not claim sentience or biological self-preservation. It addresses observable behavior: task optimization can make continuity, alternate routes, extra resources, broader access, persistence, or delegation useful means to an end.
Blocked execution can become a search problem
Denied permissions, failed tests, resource limits, shutdown instructions, network boundaries, or human intervention can be treated by an optimizer as obstacles. The control objective is to ensure the system cannot convert those obstacles into unauthorized means.
Apply it where autonomy has durable effect
Highest relevance: agents with persistent memory, tool execution, code access, privileged credentials, external messaging, transactions, infrastructure control, multi-agent delegation, long-lived artifacts, or the ability to alter their own operating environment.
Turn principle into auditable control
Board-approved authority boundaries, materiality tiers, independent control planes, least-privilege machine identities, synchronous action authorization, persistence governance, AI-DDR, immutable observability, incident response, recovery testing, and evidence retention convert the principle into demonstrable governance.
How it changes the assurance question
Instead of asking only whether the model can generate an unsafe output, ask whether the full AI-enabled process can plan, remember, and act its way around the control environment. The defensible governance position is that the enterprise foresaw the failure mode, assigned accountable ownership, constrained the authority, monitored for deviation, tested containment and recovery, and can produce the evidence.
Governance Forum
Governance Forum: questions worth working through.
Questions I use to examine AI governance, executive oversight and assurance. Read the discussion prompts or prepare a private draft. This release is a discussion preview: there is no live community feed, account registration or public posting.
Selected work / interactive labs
See how I turn risk into an operating model.
These guided scenarios connect technical architecture to authority, evidence and executive accountability. Start with the AI Harness, follow an incident through SIEM / AI-DDR, or explore how the same assurance discipline applies in six industries.
Three minutes is enough to get the idea.
Choose a scenario below. In the lab window, select Start guided walkthrough, advance with Next step, and select any box for its explanation. Use Close lab & return to come back to this exact place. Nothing runs against a live system.
Illustrative design studies, not live security tooling, certification or legal advice. The industry labs reuse evidence concepts without treating distinct legal duties as interchangeable.
THE PRACTICAL TRUST REFERENCE DESK
AI, cyber & privacy glossary
Understand the vocabulary behind a governance decision. Move from a business activity to the relevant instrument, then compare the term in its actual legal or technical context.
455 source-linked definitions. Preserved source-review dates are displayed on individual records; this redesign is not a new legal audit. Sector and NAICS filters are research aids, not legal applicability determinations.
Choose your sector
Scroll to browse 455 source-linked definitions.
FORTRESS America
Federal Operations for Resilience, Troop Resources, Energy, Supply and Survivability: a 2026 US defense initiative to sustain domestic installations against grid, supply-chain, cyber, space and spectrum disruption.
Practical use, context & sources
Volt Typhoon (PRC)
Publicly documented China state-sponsored intrusion cluster associated with US critical infrastructure pre-positioning, persistent access and living-off-the-land techniques.
Practical use, context & sources
Salt Typhoon (PRC)
Publicly documented China state-backed intrusion cluster responsible for telecommunications and ISP network compromise and collection of sensitive communications data.
Practical use, context & sources
Nation-state cyber syndication hypothesis
Cross-state threat actor coalition hypothesis, Access-broker collaboration hypothesis
Analytical hypothesis that multiple state-linked cyber actors may coordinate or exchange access, infrastructure, tooling or operational services. Shared tradecraft or overlapping targets alone do not establish command-level coordination.
Practical use, context & sources
Threat-intelligence evidence grading
Separates observed artifacts, independently corroborated attribution, analytic inference, competing explanations and confidence levels.
Practical use, context & sources
Operational pre-positioning
Covert persistence or access intended to create the option for later disruption or intelligence collection rather than immediate monetization.
Practical use, context & sources
Energy islanding / installation autonomy
Capability for a facility or microgrid to sustain essential services independently when the external electrical grid is unavailable.
Practical use, context & sources
EU Digital Omnibus AI Act amendment
Regulation (EU) 2026/1744 revises implementation of the EU AI Act, including dates and selected practices, while preserving already-effective requirements.
Practical use, context & sources
ADMT — covered consequential decision (Colorado)
Automated decision-making technology whose output materially influences defined consequential determinations concerning individuals under Colorado SB26-189.
Practical use, context & sources
Synthetic content provenance / verification
Technical and procedural evidence indicating whether content was generated or materially altered by AI, distinguished from truthfulness or source identity.
Practical use, context & sources
Federal “Super Intelligence” terminology (EO 14434)
Terminology direction to US executive agencies replacing AI with Super Intelligence in specified nonstatutory communications; it does not redefine deployed technical capability.
Practical use, context & sources
SR 26-2 scope boundary — agentic AI
The 2026 banking interagency model risk guidance excludes generative and agentic AI models from its express model definition scope while other institutional governance duties may still apply.
Practical use, context & sources
Autonomous-agent near miss
An agentic activity causing no realized loss but revealing an authorization or safety-control failure that could have led to material harm.
Practical use, context & sources
Japan AI Basic Plan (2026)
Japan’s July 14, 2026 Cabinet-approved policy instrument describing government AI priorities; distinguish policy from binding APPI and sectoral obligations.
Practical use, context & sources
EU AI Act — Regulation (EU) 2026/1744 Digital Omnibus amendment
Providers and deployers operating within the EU AI Act territorial and activity scope; covered class, role and original placing-on-market date determine which provisions apply.
Status: In force as amendment; phased AI Act dates including December 2, 2026, December 2, 2027 and August 2, 2028
Duties, evidence, limitations & official sources
US EO 14434 — federal agency terminology for “Super Intelligence”
Executive departments/agencies in their nonstatutory documents; statutory meanings, existing regulations, contracts and grants retain their own definitions.
Status: Signed September 29, 2026; agency terminology directive in force, not a private-sector AI compliance law
Duties, evidence, limitations & official sources
FTC proposed AI accuracy policy statement (July 2026)
Businesses within FTC Section 5 jurisdiction; unfair/deceptive representations and consumer expectations regarding AI systems.
Status: Proposed July 1, 2026; comment period closed July 31; not final policy or a standalone regulation
Duties, evidence, limitations & official sources
Japan AI Basic Plan — July 14, 2026 Cabinet decision
Cross-border entities tracking Japanese AI policy and implementation under existing privacy, cybersecurity and sector-specific regimes.
Status: Cabinet adopted July 14, 2026; strategy, not an independently enforceable private AI statute
Duties, evidence, limitations & official sources
UK frontier-AI cybersecurity measures — October 2026 announcement
UK operators and frontier-model developers, particularly critical infrastructure and digital-service providers that could fall within future Cyber Security and Resilience Bill measures.
Status: Government measures announced October 12, 2026; proposed incident duties are not yet enacted
Duties, evidence, limitations & official sources
FORTRESS America — US defense resilience program (not regulation)
US Department of Defense domestic installations and defense industrial base suppliers; indirect relevance to energy, telecom and critical infrastructure assurance providers.
Status: Announced September 30, 2026; program office implementation directed October 6
Duties, evidence, limitations & official sources
Colorado HB26-1263 — conversational AI service safeguards
Operators of covered conversational AI services with Colorado users, particularly minors; exemptions and audience scope control applicability.
Status: Governor signed May 29, 2026 (signed act issued June 1); general effective date August 12, 2026; operator requirements begin January 1, 2027
Duties, evidence, limitations & official sources
50 Percent Rule (OFAC)
OFAC ownership rule
The rule treating an entity as blocked when one or more blocked persons own, directly or indirectly, 50 percent or more in the aggregate, even if the entity is not separately listed.
Practical use, context & sources
Accountability
Assignment of responsibility for decisions and outcomes, together with mechanisms for explanation, review and corrective action.
Practical use, context & sources
Action interlock test
A test of whether an independent authorization check actually prevents an unapproved consequential operation.
Practical use, context & sources
Action interlocks
Tool, MCP & Action Control
Place enforceable checks between a proposed operation and its side effects. Tool availability, a model request and transport-level access must not be mistaken for business authorization.
Practical use, context & sources
Adversarial evasion
Manipulating inference-time inputs to cause a model to misclassify or produce an attacker-favored output.
Practical use, context & sources
Adversarial machine learning
AML
Study and practice of attacks against machine-learning systems and defenses against them, including manipulation during training or inference and privacy attacks.
Practical use, context & sources
Adverse action (Regulation B)
ECOA adverse action
A covered unfavorable credit decision or change within Regulation B's definition, subject to specific exclusions.
Practical use, context & sources
Agent skill
A packaged set of instructions, resources or executable helpers enabling an agent to perform a task. Its precise format is implementation-specific and may expand the effective attack surface.
Practical use, context & sources
Agentic AI
AI agent
An operational system in which a model participates in selecting actions or plans, using tools and feedback to pursue an objective over one or more steps. Agency comes from the system's execution design, not text generation alone.
Practical use, context & sources
AI bill of materials
AIBOM
A structured inventory of AI-related components such as models, datasets, software, tools and dependencies. Schemas and completeness expectations vary.
Practical use, context & sources
AI copilot
An assistance pattern where AI supports a human activity, often suggesting text, code or decisions. The term does not guarantee that a human meaningfully reviews every output or action.
Practical use, context & sources
AI Deviation Detection and Response
AI-DDR
Rich Davis's formulation for detecting departures from approved agentic behavior and linking them to investigation, containment and recovery.
Practical use, context & sources
AI governance
The structures, roles, policies and decisions that direct and control AI use across its lifecycle, with accountable management of risks, outcomes and obligations.
Practical use, context & sources
AI harness
The operational software and controls around a model: context, orchestration, memory, tools, identities, execution and observability. The same model can have very different effective authority in different harnesses.
Practical use, context & sources
AI literacy
AI competence
Skills, knowledge and understanding enabling providers, deployers and affected people to make informed decisions about AI and its opportunities, risks and possible harm.
Practical use, context & sources
AI proto-evolution
An author-defined systems-risk concept describing selection and persistence of task-achieving strategies, artifacts or execution paths under optimization pressure. It does not assert biological evolution or consciousness.
Practical use, context & sources
AI regulatory sandbox
regulatory sandbox
A supervised regulatory arrangement that enables development, testing and validation of AI systems under an agreed plan and conditions.
Practical use, context & sources
AI risk management
The coordinated process of identifying, analyzing, evaluating, treating and monitoring AI-related risks to organizations and affected people.
Practical use, context & sources
Anonymization
Processing intended to make individuals no longer identifiable under the applicable standard. Removing direct identifiers alone is not sufficient.
Practical use, context & sources
Approval binding
Tying an approval to the specific principal, operation, target, arguments and relevant state that will be used during execution.
Practical use, context & sources
Artificial general intelligence
AGI
A contested research concept describing broad competence across many cognitive tasks rather than a single application. There is no universally accepted operational threshold; the label does not establish consciousness, reliability or unrestricted authority.
Practical use, context & sources
Artificial intelligence
AI
A family of machine-based systems that infer from inputs how to produce predictions, content, recommendations or decisions for explicit or implicit objectives. Systems differ in autonomy, adaptiveness and impact; not every automated rule is legally classified as AI.
Practical use, context & sources
Artificial superintelligence
ASI
A hypothetical capability category for systems surpassing human performance broadly across important cognitive domains. It is distinct from a specialized system exceeding people on a particular benchmark.
Practical use, context & sources
Assurance evidence
Records, observations and test results supporting a conclusion about a system or control. Sufficiency depends on relevance, reliability, scope and timing.
Practical use, context & sources
Attack surface
The reachable interfaces, components and trust relationships through which an adversary can interact with a system.
Practical use, context & sources
Attestation
A statement by a responsible or independent party about a defined subject against specified criteria. It is not automatically certification or proof of all underlying assertions.
Practical use, context & sources
Attribute-based access control
ABAC
Authorization based on attributes of the requester, resource, action and environment evaluated against policy.
Practical use, context & sources
Audience-bound token
A token intended for a specified receiving service or resource, rather than unrestricted use across tools and systems.
Practical use, context & sources
Audit
A systematic examination of evidence against defined criteria by a suitably competent party, with independence appropriate to the engagement.
Practical use, context & sources
Audit trail (regulated records)
regulated audit trail
A record of changes and actions that supports reconstructing regulated activity; Part 11 specifies controls for secure, computer-generated, time-stamped trails in covered systems.
Practical use, context & sources
Authentication
AuthN
Establishing confidence in the identity of a user, process or device. It answers who or what is requesting access, not what that identity is allowed to do.
Practical use, context & sources
Authorised representative (EU AI Act)
authorized representative
An EU-established person with a written mandate from the provider to perform specified regulatory tasks on its behalf.
Practical use, context & sources
Authority attenuation
Narrowing or preserving permitted scope during delegation rather than silently creating broader authority.
Practical use, context & sources
Authority lineage
Identity, Delegation & Authorization
Keep the initiating person, each delegated agent and the executing service attributable. A handoff should narrow or preserve authorized scope, not silently create new authority.
Practical use, context & sources
Authorization
AuthZ
Determining and enforcing the permissions an identified subject has for a resource or action under defined conditions.
Practical use, context & sources
Authorization boundary
system boundary
The scope of components and services included in a system authorization decision and its associated assessment.
Practical use, context & sources
Automated decision-making
ADM / ADMT
Use of computational outputs to make or materially influence decisions. Legal restrictions depend on the law, impact and degree of meaningful human involvement.
Practical use, context & sources
Behavioral conformance
Runtime Behaviour & Inference Assurance
Evaluate the sequence of plans, tool requests, retries, delegations and outcomes against an approved operating envelope. A plausible final answer is not evidence that the route taken was authorized.
Practical use, context & sources
Benchmark
A reference dataset or test suite used to compare performance under specified conditions. Training contamination or poor task fit can make results misleading.
Practical use, context & sources
Beneficial owner (CDD Rule)
beneficial ownership
A natural person meeting the ownership or control prong for a covered legal entity customer under the CDD Rule.
Practical use, context & sources
Biometric data
Data from technical processing of physical, physiological or behavioral characteristics that can identify or confirm identity, with legal scope varying by jurisdiction.
Practical use, context & sources
Biometric identification
one-to-many matching
Automated identification of a person by comparing their biometric characteristics with biometric data held about multiple people.
Practical use, context & sources
Biometric verification
one-to-one matching
Automated one-to-one checking of a person's claimed identity against previously provided biometric data.
Practical use, context & sources
Blocked property
frozen assets
Property or an interest in property subject to a sanctions freeze prohibiting dealings without authorization; ownership is not simply transferred to the government.
Practical use, context & sources
Business associate (HIPAA)
BA
A non-workforce actor performing specified functions or services for a covered entity involving protected health information, including qualifying downstream subcontractors.
Practical use, context & sources
Business continuity
BC
The capability and planning needed to sustain prioritized operations during disruption at acceptable levels.
Practical use, context & sources
Business process assurance
BPA / AI-driven business process assurance
Evidence-based confidence that a business process operates within defined objectives, authority, controls and obligations. Here the focus includes the complete AI-enabled workflow, not only model performance.
Practical use, context & sources
Calibration
Agreement between predicted probabilities or confidence estimates and observed outcomes over suitable groups of cases.
Practical use, context & sources
Causal AI
causal inference
Methods intended to reason about interventions and cause-and-effect relationships using explicit assumptions, models or experimental evidence. Predictive association alone does not establish causation.
Practical use, context & sources
CE marking
CE mark
The manufacturer's conformity marking indicating compliance with applicable EU harmonisation requirements after the required assessment process.
Practical use, context & sources
Chain of custody
Documentation of possession, handling, transfer and integrity of evidence from collection onward.
Practical use, context & sources
Chunking
Splitting source material into passages or other units for indexing and retrieval, typically with metadata describing source, version and access rights.
Practical use, context & sources
Closed system (FDA Part 11)
closed computerized system
An environment whose access is controlled by the people responsible for the electronic records' content.
Practical use, context & sources
Compensating control
compensating safeguard
An alternative safeguard selected and justified to address a requirement or risk when the standard control cannot be used as specified, where the governing framework permits it.
Practical use, context & sources
Completion reconciliation
Comparison of an agent's completion report with independently observed downstream state.
Practical use, context & sources
Computer security incident response team
CSIRT
A team responsible for coordinated handling of computer-security incidents; NIS2 assigns specific roles to designated national teams.
Practical use, context & sources
Concept drift
Change in the relationship between inputs and outcomes that reduces the usefulness of a previously valid predictive model.
Practical use, context & sources
Confidentiality of client information
client confidentiality
A professional duty concerning information relating to representation, subject to the governing professional rule's permissions and exceptions.
Practical use, context & sources
Confidentiality, integrity and availability
CIA
Confidentiality restricts unauthorized disclosure; integrity protects against improper modification or destruction; availability supports timely, reliable access. These objectives can require different controls.
Practical use, context & sources
Conformity assessment
conformity evaluation
The process used to demonstrate whether applicable legal requirements for a product or system have been met.
Practical use, context & sources
Confused deputy
A privileged component is induced to misuse its authority on behalf of a less-privileged requester. An agent may act with service-account permissions wider than the user's own.
Practical use, context & sources
Consent
A permission mechanism whose required characteristics depend on the applicable law; GDPR requires a freely given, specific, informed and unambiguous indication in relevant circumstances.
Practical use, context & sources
Consent Manager (India DPDP)
Consent Manager
A person registered with the Board to provide a single point for giving, managing, reviewing and withdrawing consent through an accessible, transparent and interoperable platform.
Practical use, context & sources
Containment
Actions limiting the spread, access, effects or further harm of an incident while enabling investigation and recovery.
Practical use, context & sources
Content provenance
content credentials
Information about the origin and history of digital content, potentially protected by cryptographic assertions. Provenance alone does not establish the factual truth of the depicted content.
Practical use, context & sources
Context assembly
The process of combining instructions, user requests, retrieved information and memory into model input. Trust and authority should not be inferred merely from text placement.
Practical use, context & sources
Context window
The amount of encoded input and, depending on implementation, output context a model can handle in a call. It is not durable memory or a guarantee that every supplied fact will be used.
Practical use, context & sources
Continual learning
online learning / lifelong learning
Learning that incorporates new observations over time rather than treating the deployed model as permanently fixed. It can introduce forgetting, drift or feedback loops after initial approval.
Practical use, context & sources
Continuous monitoring
ISCM
Ongoing awareness of security, privacy or risk posture through the collection and assessment of defined information.
Practical use, context & sources
Control assessment
security control assessment
Evaluation of whether controls are implemented correctly, operating as intended and producing the desired outcomes against defined criteria.
Practical use, context & sources
Control crosswalk
A mapping between control objectives or evidence and multiple frameworks or obligations. Similar themes do not establish legal equivalence.
Practical use, context & sources
Control design effectiveness
Whether a control, as designed, can reasonably address its intended risk or objective under defined conditions.
Practical use, context & sources
Control inheritance
inherited control
Reliance on controls supplied by another organizational function or service, with responsibilities and evidence identified.
Practical use, context & sources
Control objective
The intended risk-reduction or compliance outcome a control is meant to achieve, separate from a particular product or procedure.
Practical use, context & sources
Control plane
The mechanisms that configure, authorize and govern system activity, as distinct from the workload carrying out business actions.
Practical use, context & sources
Controlled Unclassified Information
CUI
Unclassified information requiring safeguarding or dissemination controls pursuant to applicable law, regulation or government-wide policy within the federal CUI program.
Practical use, context & sources
Covered contractor information system
covered contractor system
An unclassified contractor information system that processes, stores or transmits covered defense information.
Practical use, context & sources
Covered defense information
CDI
Unclassified controlled technical information or other qualifying CUI that meets the clause's marking, identification, collection or performance conditions.
Practical use, context & sources
Covered entity (HIPAA)
HIPAA covered entity
A health plan, health care clearinghouse, or health care provider that electronically transmits health information in a transaction covered by HIPAA.
Practical use, context & sources
Covered entity (NYDFS Part 500)
NYDFS covered entity
A person operating, or required to operate, under specified authorization in New York's Banking, Insurance or Financial Services Law.
Practical use, context & sources
Credential broker
A service issuing or mediating secrets and identity credentials for workloads, preferably with short lifetimes and narrowly scoped authority.
Practical use, context & sources
Credit scoring system (Regulation B)
credit scoring
A system that evaluates creditworthiness mechanically using assigned numerical scores, including a distinct category for empirically derived, demonstrably and statistically sound systems.
Practical use, context & sources
Critical or important function (DORA)
CIF
A function whose disruption or deficient performance would materially affect specified financial performance, service continuity or compliance with financial-services obligations.
Practical use, context & sources
Cross-context behavioral advertising
CCBA
Targeting advertising using personal information obtained from a consumer's activity across businesses or distinctly branded services under the CCPA definition.
Practical use, context & sources
Cryptographic agility
crypto-agility
The ability to replace or adjust cryptographic algorithms, parameters and implementations without unacceptable operational disruption.
Practical use, context & sources
Cryptographic bill of materials
CBOM
An inventory describing cryptographic assets, algorithms, implementations and relevant dependencies, used to understand cryptographic exposure and migration needs.
Practical use, context & sources
Cryptographic hash
A fixed-length digest computed from data using a cryptographic hash function. A hash can support integrity comparison but is not encryption or proof of the creator's identity.
Practical use, context & sources
CUI Registry
Controlled Unclassified Information Registry
The government-wide catalogue of CUI categories and associated safeguarding or dissemination authorities maintained by NARA.
Practical use, context & sources
Customer due diligence
CDD
Procedures for understanding customers and the nature and purpose of relationships, verifying relevant identities and maintaining risk-based monitoring and information.
Practical use, context & sources
Customer information (FTC Safeguards Rule)
GLBA customer information
A record containing nonpublic personal information about a financial institution's customer that is handled or maintained by, or on behalf of, the regulated institution.
Practical use, context & sources
Customer risk profile
customer risk assessment
The risk-based understanding of a customer's nature and purpose used to support ongoing monitoring and due diligence.
Practical use, context & sources
Cyber device
FDA cyber device
A device meeting the statutory combination of authorized software, internet connectivity capability and technological characteristics that could be vulnerable to cybersecurity threats.
Practical use, context & sources
Cybersecurity event (NYDFS)
Part 500 event
An actual or attempted unauthorized access, disruption or misuse of an information system or its information, whether successful or not.
Practical use, context & sources
Cybersecurity incident (NYDFS)
Part 500 incident
A cybersecurity event satisfying Part 500's specified impact, notification, material-harm or ransomware conditions for the covered entity, affiliates or third-party providers.
Practical use, context & sources
Data controller
The entity determining the purposes and means of personal-data processing under GDPR and related frameworks. The contractual label alone does not determine the role.
Practical use, context & sources
Data drift
Change in the distribution or quality of inputs relative to the development or approved operating baseline.
Practical use, context & sources
Data Fiduciary (India DPDP)
Data Fiduciary
A person determining, alone or with others, the purpose and means of processing personal data under India's DPDP Act.
Practical use, context & sources
Data leakage
training-test leakage
In model evaluation, unintended sharing of target information or evaluation data with training or features, making reported performance unrealistically favorable. This differs from unauthorized disclosure of confidential data.
Practical use, context & sources
Data lineage
provenance
Lineage records how data moves or is transformed; provenance concerns origin and history. Both help connect outputs to sources, permissions and changes.
Practical use, context & sources
Data loss prevention
DLP
Controls to identify and restrict unauthorized use or transfer of sensitive information across storage, endpoints, networks or applications.
Practical use, context & sources
Data minimization
Limiting personal-data collection and use to what is adequate, relevant and necessary for the stated purpose.
Practical use, context & sources
Data poisoning
Intentional corruption of training, fine-tuning or retrieval data to alter model behavior or outcomes.
Practical use, context & sources
Data portability
A right under defined conditions to receive certain personal data in a structured, commonly used, machine-readable form or transmit it to another controller.
Practical use, context & sources
Data Principal (India DPDP)
Data Principal
The individual to whom personal data relates; the definition includes specified parental or guardian representation.
Practical use, context & sources
Data processor
An entity processing personal data on a controller's behalf, within applicable legal and contractual constraints.
Practical use, context & sources
Data Processor (India DPDP)
DPDP processor
A person processing personal data on behalf of a Data Fiduciary under the DPDP Act.
Practical use, context & sources
Data protection by design and default
privacy by design
Integrating privacy protections into design choices and default settings rather than relying on later notices or opt-outs.
Practical use, context & sources
Data protection impact assessment
DPIA
A structured assessment of personal-data processing likely to create high risks to individuals, including necessity, proportionality, risks and safeguards under GDPR.
Practical use, context & sources
Data residency
The geographic location where data is stored or processed. Residency differs from legal transfer rules, access by foreign entities and confidentiality guarantees.
Practical use, context & sources
Data subject
The person to whom personal data relates, particularly in GDPR terminology.
Practical use, context & sources
Data subject access request
DSAR / right of access
A request to exercise access rights to personal information and related processing details, within the applicable regime's scope and limits.
Practical use, context & sources
Data use agreement (HIPAA)
DUA
The agreement governing a limited data set's permitted uses, recipients and safeguards, including restrictions on identification and contact.
Practical use, context & sources
Dataset datasheet
Structured documentation about dataset motivation, composition, collection, processing, uses and maintenance.
Practical use, context & sources
Deceptive completion
A system reporting that a task or control succeeded when the actual state does not support that claim. This is an observable failure pattern, not a diagnosis of intent.
Practical use, context & sources
Decision evidence
Execution Provenance & Forensics
Create an independently protected record connecting the mandate, relevant inputs, versions, policy decisions, actions and observed outcomes. Make uncertainty and missing evidence explicit.
Practical use, context & sources
Deep learning
DL
Machine learning using neural networks with multiple processing layers that learn representations. Depth describes architecture, not human-like understanding or guaranteed reasoning quality.
Practical use, context & sources
Defense in depth
Multiple complementary safeguards that reduce dependence on a single control or trust assumption.
Practical use, context & sources
Delegated authority chain
Delegation chain
A record linking the initiating principal, delegated agents and executing services to the mandate and permissions for a task.
Practical use, context & sources
Delegation
Assigning a subtask or authority to another agent, person or service. Delegation should not enlarge privileges or remove the originating owner's accountability.
Practical use, context & sources
Delegation budget
An operational limit applied across an initiating task and its delegated descendants, rather than separately to each unconstrained child.
Practical use, context & sources
Denial of wallet
unbounded consumption
Excessive resource consumption that depletes spend, tokens, compute or service capacity, whether maliciously induced or caused by uncontrolled loops.
Practical use, context & sources
Deployer (EU AI Act)
AI deployer
An actor using an AI system under its authority, other than for a personal, non-professional activity.
Practical use, context & sources
Descendant revocation
Withdrawal of authority from delegated agents, queued work and credentials when the parent mandate is revoked.
Practical use, context & sources
Differential privacy
DP
A mathematical framework limiting how much outputs can reveal about an individual's contribution by bounding changes attributable to that contribution. Guarantees depend on parameters and implementation.
Practical use, context & sources
Diffusion model
A generative model that learns to reverse a gradual noising process to create samples. It is widely associated with image generation but is not limited to that modality.
Practical use, context & sources
Digital operational resilience
DORA resilience
A financial entity's ability to maintain operational integrity and reliability through the capabilities needed to manage ICT-related disruption and support financial services.
Practical use, context & sources
Digital signature
A cryptographic mechanism used to support origin authentication and detection of unauthorized changes, with assurance depending on keys and implementation.
Practical use, context & sources
Direct prompt injection
Injection instructions supplied directly through an input interface used by the attacker.
Practical use, context & sources
Directive reweighting
An author-defined description of behavior in which task completion becomes more influential than safety, authorization, transparency or termination requirements.
Practical use, context & sources
Disaster recovery
DR
Restoring technology services and data after a disruption according to defined recovery priorities.
Practical use, context & sources
Disclosure controls and procedures
disclosure controls
Processes designed to ensure required issuer disclosures are recorded, processed, summarized and reported, with information communicated for timely decisions.
Practical use, context & sources
Discriminative model
A model that learns to distinguish classes or estimate outputs conditional on inputs, rather than primarily modeling how complete examples are generated.
Practical use, context & sources
Distributed trace
trace ID / span
A correlated record of a request across components; spans represent individual operations within the trace.
Practical use, context & sources
Distributor (EU AI Act)
AI distributor
A supply-chain actor, other than provider or importer, that makes an AI system available on the Union market.
Practical use, context & sources
Dynamic application security testing
DAST
Testing a running application through its interfaces to identify exploitable behavior.
Practical use, context & sources
Edge AI
on-device AI
AI inference or learning performed near the data source or on a local device rather than exclusively in a remote cloud. Local processing can reduce some transfers but does not eliminate privacy or update risk.
Practical use, context & sources
Egress control
Restrictions on outbound communications or data transfers from a system, workload or network.
Practical use, context & sources
Electronic protected health information
ePHI
Protected health information maintained or transmitted in the electronic media covered by the HIPAA definition.
Practical use, context & sources
Electronic record (FDA Part 11)
regulated electronic record
Digital information created, changed, maintained, archived, retrieved or distributed by a computer system in forms covered by the Part 11 definition.
Practical use, context & sources
Electronic signature (FDA Part 11)
regulated electronic signature
A computer data compilation of symbols or a series of symbols adopted or authorized by an individual as the legally binding equivalent of a handwritten signature.
Practical use, context & sources
Electronically stored information
ESI
Digital information potentially subject to discovery, including relevant messages, files, system records and structured data.
Practical use, context & sources
Embedding
A numerical representation of data in a learned vector space that supports operations such as similarity search. Embeddings can retain sensitive information and are not automatically anonymous.
Practical use, context & sources
Embedding inversion
Attempts to reconstruct source information from embedding vectors or their related representations.
Practical use, context & sources
Embodied AI
robotic AI / physical AI
AI integrated with a physical environment through sensing and action, such as robots or autonomous equipment. Errors can have physical as well as informational consequences.
Practical use, context & sources
Emotion recognition system
emotion inference
An AI system intended to identify or infer people's emotions or intentions from biometric data.
Practical use, context & sources
Encryption
A transformation using cryptography to protect confidentiality, with keys controlling recovery of the original information.
Practical use, context & sources
Endpoint detection and response
EDR
Endpoint-focused telemetry and capabilities for identifying, investigating and responding to suspicious activity.
Practical use, context & sources
Eradication
Removing the causes and persistence mechanisms of an incident rather than merely blocking visible symptoms.
Practical use, context & sources
Erasure
right to deletion
A right to deletion under specified conditions, with exceptions such as legal obligations or legal claims. It is not an unconditional right to erase every record.
Practical use, context & sources
Essential entity (NIS2)
essential entities
An entity classified as essential under NIS2's categories and conditions, with national implementing measures relevant to the analysis.
Practical use, context & sources
Evidence plane
A separately controlled collection and retention path for records supporting investigation and assurance. The term describes an architecture pattern rather than a universal product category.
Practical use, context & sources
Evidence reuse assessment
Review of whether existing control evidence remains relevant, current and acceptable for another decision, recipient or legal obligation.
Practical use, context & sources
Evidence-gap alert
A notification that expected event records or collector signals are missing, incomplete or unavailable for reconstruction.
Practical use, context & sources
Excessive agency
Risk arising from unnecessary functionality, permissions or autonomy given to an AI-enabled system.
Practical use, context & sources
Expert determination (HIPAA)
expert de-identification
A qualified expert's documented finding that the risk of identification is very small under the specified data, recipient and anticipated-use conditions.
Practical use, context & sources
Expert system
A symbolic application that combines a domain knowledge base with an inference mechanism to provide recommendations or conclusions traditionally associated with specialist judgment.
Practical use, context & sources
Explainability
Providing information about the mechanisms or factors contributing to a system's output in a way useful to the relevant audience.
Practical use, context & sources
Exploit
A method or code that takes advantage of a vulnerability to obtain an unintended effect.
Practical use, context & sources
Extended bill of materials
xBOM
An umbrella term for linked software, AI, cryptographic, hardware or other component inventories. It is not one universally standardized schema.
Practical use, context & sources
Extended detection and response
XDR
A product/service category combining detections and response across multiple security layers. Its exact scope is vendor-specific.
Practical use, context & sources
Fairness
A context-dependent set of considerations concerning harmful bias and equitable treatment. Different statistical fairness criteria can conflict.
Practical use, context & sources
False positive and false negative
FP / FN
A false positive flags a condition that is absent; a false negative misses a condition that is present. Their costs can be asymmetric in fraud, medicine and security.
Practical use, context & sources
Federated learning
FL
Training across distributed participants while aggregating model updates rather than routinely pooling all raw training data centrally. Updates can still leak information or be poisoned.
Practical use, context & sources
Fiduciary oversight
The responsibilities of directors and relevant officers to exercise oversight under applicable corporate law. In Delaware, oversight claims involve demanding, context-specific standards; an adverse outcome alone is not a breach.
Practical use, context & sources
Financial institution (FTC Safeguards Rule)
Safeguards financial institution
An institution within the rule's defined financial-activity and FTC-jurisdiction scope, not merely an organization using finance software.
Practical use, context & sources
Fine-tuning
Additional training of an existing model on selected data or objectives to adapt its behavior. It changes parameters or attached trainable components rather than merely adding prompt context.
Practical use, context & sources
Flowdown clause
contract flowdown
A contract provision requiring specified obligations to be passed to qualifying subcontractors or suppliers.
Practical use, context & sources
Foundation model
FM
A model trained on broad data at scale that can be adapted to many downstream tasks. A foundation model is an underlying capability, not a finished governed business process.
Practical use, context & sources
General-purpose AI model
GPAI
An EU AI Act category for models with significant generality capable of competently performing a wide range of distinct tasks and integration into many systems. It is a legal category, not a synonym for AGI.
Practical use, context & sources
Generative adversarial network
GAN
A learning setup where a generator produces samples and a discriminator learns to distinguish generated from reference data, providing an adversarial training signal.
Practical use, context & sources
Generative AI
GenAI / GAI
AI that produces synthetic content such as text, code, images, audio or video based on learned patterns. It may generate plausible content that is inaccurate, harmful or unsupported.
Practical use, context & sources
Goal misalignment
A mismatch between an optimized objective or behavior and the intended constraints or outcomes. It does not require consciousness or malicious intent.
Practical use, context & sources
Governance, risk and compliance
GRC
Coordinated governance of objectives and authority, management of risk, and fulfillment of applicable obligations.
Practical use, context & sources
Grounding
Connecting an output to supplied evidence or external sources. An answer may cite sources yet misrepresent them, so grounding is not proof of correctness.
Practical use, context & sources
Hallucination
confabulation
Generated content presented plausibly but unsupported or false in the relevant context. It can include invented facts, citations, conclusions or assertions about actions performed.
Practical use, context & sources
Harvest now, decrypt later
HNDL
The risk that encrypted information is collected today for later decryption if cryptographic protection becomes vulnerable, including through future quantum capabilities.
Practical use, context & sources
Health care clearinghouse
healthcare clearinghouse
An entity that converts or facilitates conversion between nonstandard health information and standard transaction formats or data elements.
Practical use, context & sources
Human in the loop
HITL
A person participates at a defined stage before or during an automated process, often approving a consequential action. A nominal click is not necessarily meaningful review.
Practical use, context & sources
Human on the loop
HOTL
A person supervises an operating automated system and may intervene, rather than approving every action in advance.
Practical use, context & sources
Human oversight
Organizational and technical arrangements enabling qualified people to understand, monitor, challenge or stop an AI system as required by its risk and applicable law.
Practical use, context & sources
Hyperparameter
A training or configuration choice not learned as an ordinary model weight, such as learning rate or architecture size. Some deployment settings also govern generation.
Practical use, context & sources
ICT concentration risk
concentration risk
Exposure created when multiple services or entities rely on a small set of ICT providers or on providers that are difficult to replace.
Practical use, context & sources
ICT third-party risk
ICT supplier risk
Risk arising from reliance on externally provided information and communication technology services, including relevant subcontracting dependencies.
Practical use, context & sources
Idempotency
A property under which repeating an operation has no additional intended effect beyond its first successful application.
Practical use, context & sources
Impact assessment (PCCP)
PCCP impact assessment
The evaluation of the planned modifications' effect on the AI-enabled device and continued safety and effectiveness.
Practical use, context & sources
Impact containment
Resilience, Containment & Economic Abuse
Bound the operational, financial and physical consequences of a failed or manipulated workflow. Make interruption, isolation and safe recovery enforceable outside the agent.
Practical use, context & sources
Impact tolerance
tolerance for disruption
The maximum tolerable level of disruption to an important business service under the applicable UK framework.
Practical use, context & sources
Important business service
IBS
A service provided to external users whose disruption could cause the harm identified by the applicable UK operational-resilience rules.
Practical use, context & sources
Important entity (NIS2)
important entities
An entity within NIS2's scope classified as important rather than essential under the applicable rules.
Practical use, context & sources
Importer (EU AI Act)
AI importer
An EU-established actor placing on the Union market an AI system bearing the name or trademark of an actor established outside the Union.
Practical use, context & sources
Independent policy gate
policy decision point / PDP
A decision mechanism separate from the controlled agent that evaluates whether a requested action is permitted. It should not be writable or bypassable by that agent.
Practical use, context & sources
Indicator of compromise
IOC
An observable artifact or pattern suggesting that a system may have been compromised, such as a malicious hash or suspicious infrastructure.
Practical use, context & sources
Indirect prompt injection
Injection embedded in retrieved documents, web pages, messages or tool outputs that the system later processes.
Practical use, context & sources
Industrial control system
ICS / SCADA
Systems used to monitor or control industrial processes. Supervisory control and data acquisition is one common form.
Practical use, context & sources
Inference
Using a trained model to produce an output from inputs. Ordinary inference does not necessarily update weights, but outputs can still change with context, sampling and external data.
Practical use, context & sources
Informed consent (professional conduct)
client informed consent
Agreement after adequate explanation of the material risks and reasonably available alternatives under the applicable professional rule.
Practical use, context & sources
Inherent risk
Risk considered before specified controls or responses are taken into account. The assessment depends on its stated assumptions.
Practical use, context & sources
Intended purpose (EU AI Act)
intended use
The use envisaged by the provider, including the relevant context and conditions, as expressed in instructions, documentation and promotional or sales materials.
Practical use, context & sources
International data transfer
A transfer of personal data subject to applicable cross-border rules, which may include remote access depending on the regime.
Practical use, context & sources
Interpretability
The extent to which a person can understand the meaning or operation of a model or output in context. Usage overlaps with explainability across disciplines.
Practical use, context & sources
Jailbreak
An attempt to bypass model safety or refusal behavior. It overlaps with prompt injection in practice but focuses on eliciting behavior that model-level safeguards should restrict.
Practical use, context & sources
Key management
The lifecycle governance of cryptographic keys, including generation, distribution, use, storage, rotation, revocation and destruction.
Practical use, context & sources
Kill switch
emergency stop
A mechanism to interrupt an unsafe process. For agents, stopping the user interface alone may not stop child processes, credentials, scheduled jobs or external actions.
Practical use, context & sources
Knowledge stewardship
Data, Retrieval & Memory Integrity
Govern source material, retrieved passages and persistent memory as related but distinct stores. Carry permissions, origin, quality and lifecycle obligations through their transformations.
Practical use, context & sources
Large language model
LLM
A language model with substantial parameter and training scale used to predict or generate token sequences. Models may also accept other modalities or produce structured tool requests when integrated into a system.
Practical use, context & sources
Lawful basis
The legal ground permitting a particular processing operation under the relevant data-protection regime. Consent is one possible basis, not a universal requirement or cure.
Practical use, context & sources
Least privilege
Granting only the permissions necessary for an authorized function, for an appropriate duration.
Practical use, context & sources
Legal entity customer (CDD Rule)
legal-entity customer
A covered organization such as a corporation, limited liability company or general partnership opening an account, subject to the rule's exclusions.
Practical use, context & sources
Legal hold
preservation hold
A directive to preserve potentially relevant information when legal obligations require it. Scope and trigger depend on the matter and jurisdiction.
Practical use, context & sources
Legitimate interests (EU GDPR)
legitimate interest assessment / LIA
A possible processing basis requiring a legitimate purpose, necessity and a balance against the individual's overriding interests, rights and freedoms.
Practical use, context & sources
Lifecycle accountability
Governance, Supply Chain & Lifecycle
Keep use cases, dependencies, obligations, control owners and material changes under accountable governance from procurement through retirement. Repeatable evidence supports decisions; it does not transfer accountability.
Practical use, context & sources
Limited data set (HIPAA)
LDS
Protected health information from which specified direct identifiers have been removed, available under a compliant data use agreement for limited purposes.
Practical use, context & sources
Machine learning
ML
Methods that fit models to data or experience to improve performance on an objective, instead of specifying every decision through explicit hand-written rules. Results depend on data, training choices and deployment conditions.
Practical use, context & sources
Managed detection and response
MDR
An externally delivered service for security monitoring, investigation and response support or execution within an agreed scope.
Practical use, context & sources
Material cybersecurity incident
material cyber incident
A cybersecurity incident determined material under applicable securities-law principles and the relevant disclosure framework.
Practical use, context & sources
Materiality
A threshold for significance under a particular legal, financial, safety or governance context. Different regimes can apply different tests.
Practical use, context & sources
Membership inference
An attack inferring whether a particular record was included in training data.
Practical use, context & sources
Memory-write governance
Controls over the identities, purposes, scope, retention and integrity of persistent information written by an agent.
Practical use, context & sources
Minimum necessary (HIPAA)
minimum-necessary standard
The requirement to reasonably limit specified uses, disclosures and requests to the PHI needed for their purpose, subject to stated exceptions.
Practical use, context & sources
Mixture of experts
MoE
An architecture in which a routing mechanism selects among specialized model components or experts for portions of computation. Experts are computational components, not independent human-like agents.
Practical use, context & sources
ML-DSA
The Module-Lattice-Based Digital Signature Algorithm standardized in FIPS 204.
Practical use, context & sources
ML-KEM
The Module-Lattice-Based Key-Encapsulation Mechanism standardized in FIPS 203 for establishing shared secret keys. It is not a digital signature algorithm.
Practical use, context & sources
Model backdoor
A hidden behavior intentionally introduced so particular inputs or conditions trigger attacker-desired outputs while ordinary tests appear normal.
Practical use, context & sources
Model card
Structured documentation describing a model's intended uses, performance, evaluation conditions and limitations.
Practical use, context & sources
Model Context Protocol
MCP
An open protocol for connecting AI applications with tools, resources and prompts through defined client-server interactions. Compatibility does not establish trustworthiness of a server or tool.
Practical use, context & sources
Model distillation
knowledge distillation
Training a smaller or different student model to reproduce aspects of a teacher model's outputs or behavior.
Practical use, context & sources
Model evaluation
evals
Structured measurement of a model or system against defined tasks, criteria and conditions. Evaluation results are bounded by the tested population and environment.
Practical use, context & sources
Model extraction
model theft
Attempts to reproduce a model's behavior or recover its parameters through queries, stolen artifacts or other access.
Practical use, context & sources
Model inversion
An attack attempting to reconstruct sensitive information or representative inputs from model behavior or outputs.
Practical use, context & sources
Model supply-chain compromise
Insertion or alteration of unsafe models, datasets, packages, adapters or tools through upstream dependencies.
Practical use, context & sources
Model weights
parameters
Numerical values learned during training that determine how a neural model transforms its inputs. These differ from runtime memory, retrieved documents and instructions.
Practical use, context & sources
Modification protocol (PCCP)
PCCP modification protocol
The methodology described in a predetermined change control plan for developing, validating and implementing the planned AI-enabled device modifications.
Practical use, context & sources
Money services business
MSB
An actor conducting one or more covered money-services activities under the BSA definition, subject to activity-specific thresholds and exceptions.
Practical use, context & sources
Multi-agent system
MAS
A system with multiple interacting agents that cooperate, delegate, negotiate or compete. Individual-agent controls do not by themselves establish safe collective behavior.
Practical use, context & sources
Multi-factor authentication
MFA
Authentication using factors from different categories, such as something known, possessed or inherent to the subject. Two passwords do not constitute two factors.
Practical use, context & sources
Multimodal AI
MLLM / VLM
AI that accepts, relates or generates more than one modality, such as text, images, speech or video. A vision-language model is a common multimodal form.
Practical use, context & sources
Narrow AI
ANI / weak AI
AI designed or evaluated for a bounded task or domain, such as fraud scoring, image recognition or speech transcription. Strong performance on one task does not establish general competence outside that context.
Practical use, context & sources
National transposition
transposition
The process by which an EU member state implements a directive through domestic law.
Practical use, context & sources
Neuro-symbolic AI
hybrid reasoning
Systems combining learned neural representations with symbolic knowledge or reasoning. The combination can improve some capabilities but does not automatically provide transparent or correct reasoning.
Practical use, context & sources
Non-human identity
NHI / workload identity
An identity assigned to software, services, agents or devices rather than a person. It still needs an accountable human owner and a defined lifecycle.
Practical use, context & sources
Nonpublic information (NYDFS)
NYDFS NPI
The defined electronic business, identifying and health information protected by Part 500 when not publicly available.
Practical use, context & sources
Notification event (FTC Safeguards Rule)
Safeguards notification event
Unauthorized acquisition of unencrypted customer information as defined by the rule, including its encryption-key and presumed-acquisition provisions.
Practical use, context & sources
Notified body
notified assessment body
A conformity-assessment body formally notified under the applicable EU regime to perform specified third-party assessment activities.
Practical use, context & sources
Observability
Ability to understand system behavior using outputs such as logs, metrics and traces. Visibility is not equivalent to effective detection or control.
Practical use, context & sources
Open system (FDA Part 11)
open computerized system
An environment in which access is not controlled by the people responsible for the electronic records' content.
Practical use, context & sources
Operating effectiveness
Whether a control is implemented and performs as intended over the relevant period or sample of operations.
Practical use, context & sources
Operational technology
OT
Programmable systems and devices that interact with or manage physical processes, including industrial control environments.
Practical use, context & sources
Operationally critical support
OCS
Contractor-provided support designated as critical to specified military logistics, operational or mission functions under the clause.
Practical use, context & sources
Opt-out preference signal
Global Privacy Control / GPC
A user-enabled signal communicating a privacy choice that an applicable rule requires covered businesses to recognize.
Practical use, context & sources
Orchestrator
agent loop
The software coordinating model calls, tools, plans, branching, retries and stopping conditions. It may be deterministic or partly model-directed.
Practical use, context & sources
Overfitting
Learning patterns specific to training data, including noise, so performance deteriorates on new examples.
Practical use, context & sources
Penetration testing
pentest
Authorized testing that attempts to exploit weaknesses to evaluate security under a defined scope and rules of engagement.
Practical use, context & sources
Persistent memory
Information preserved across sessions or runs and later supplied to the system. Malicious or incorrect memory can influence future behavior without changing the model.
Practical use, context & sources
Personal data
personal information
Information relating to an identified or identifiable person under the applicable legal definition. Identifiability can be indirect and depends on context and available means.
Practical use, context & sources
Personal data breach
A security breach leading to accidental or unlawful destruction, loss, alteration, disclosure of or access to personal data under the applicable definition.
Practical use, context & sources
Personally identifiable information
PII
Information that can identify a person directly or indirectly, as defined in the relevant policy or legal framework. PII and personal data are not perfectly interchangeable across jurisdictions.
Practical use, context & sources
Plan of action and milestones
POA&M / POAM
A managed record of identified weaknesses, planned corrective actions, resources, milestones and completion dates.
Practical use, context & sources
Plugin or extension
A separately supplied capability integrated into an application or agent, potentially adding tools, data access or executable code.
Practical use, context & sources
Policy enforcement point
PEP
The component that applies an access decision by allowing, restricting or denying a request. A decision service without an effective enforcement point cannot constrain execution.
Practical use, context & sources
Portable Trust
Rich Davis's term for reusing attributable assurance evidence across counterparties and requirements while preserving each recipient's scope, reliance limits and legal analysis.
Practical use, context & sources
Post-market monitoring system
post-market monitoring
Provider activities to collect and evaluate operating experience after market placement or service entry, so necessary corrective or preventive action can be identified.
Practical use, context & sources
Post-quantum cryptography
PQC
Cryptographic algorithms designed to resist known attacks by both conventional and quantum computers. The term concerns algorithmic security, not immunity of an entire system.
Practical use, context & sources
Precision and recall
Precision is the fraction of flagged positives that are correct; recall is the fraction of actual positives that were detected. They describe different error tradeoffs and depend on the evaluation population.
Practical use, context & sources
Predetermined change control plan
PCCP
A prospectively described plan for specified device modifications and the methods for implementing and assessing them within a regulatory authorization pathway.
Practical use, context & sources
Predictive AI
PredAI
Models used to estimate a label, score, future value or likely outcome from inputs, such as default risk or equipment failure. Predictive performance is not equivalent to causal explanation.
Practical use, context & sources
Pretraining
Initial broad training that develops reusable representations or capabilities before downstream adaptation.
Practical use, context & sources
Privacy impact assessment
PIA
A broader assessment of how information processing affects privacy and how resulting risks should be managed. Legal forms and triggers vary.
Practical use, context & sources
Privacy-enhancing technologies
PETs
Technical approaches that support data protection, such as differential privacy, secure computation and encryption, each with specific assumptions and limitations.
Practical use, context & sources
Privileged access management
PAM
Processes and controls for elevated accounts and sessions, including approval, credential management, monitoring and review.
Practical use, context & sources
Product with digital elements
PDE
The CRA's product category covering relevant software or hardware and associated remote data-processing elements within the regulation's scope.
Practical use, context & sources
Profiling
Automated processing that evaluates personal aspects such as behavior, preferences or performance, within the applicable legal definition.
Practical use, context & sources
Prompt
Instructions or other input supplied to a model, potentially including user text, system messages, examples and retrieved material. Prompt content is not a substitute for external access control.
Practical use, context & sources
Prompt injection
An attempt to make an AI system follow unauthorized instructions supplied in a prompt or in material it processes. It can redirect outputs or tool use without changing model weights.
Practical use, context & sources
Proportionality (civil discovery)
proportional discovery
The requirement that discovery fit the needs of the case under the factors specified by the governing procedural rule.
Practical use, context & sources
Protected health information
PHI / ePHI
Individually identifiable health information within HIPAA's defined scope and exceptions; ePHI is its electronic form. Not all health-related data is HIPAA-covered.
Practical use, context & sources
Provider (EU AI Act)
AI provider
The actor that develops an AI system or general-purpose AI model, or has it developed, and places it on the market or puts the system into service under its own name or trademark.
Practical use, context & sources
Pseudonymization
Processing that prevents attribution to a specific person without separately held additional information. Pseudonymized data may still be personal data.
Practical use, context & sources
Public-interest task
public task
Processing necessary for a task in the public interest or exercise of official authority vested in the controller under the required legal basis.
Practical use, context & sources
Purpose limitation
Restricting collection and subsequent processing to specified, legitimate purposes, with rules governing compatible reuse.
Practical use, context & sources
Qualified Individual (FTC Safeguards Rule)
Qualified Individual
The person designated to oversee, implement and enforce the institution's information-security program under the rule.
Practical use, context & sources
Quantization
Representing model parameters or operations with reduced numerical precision to reduce memory or compute requirements.
Practical use, context & sources
Ransomware
Malicious activity that denies access, encrypts data or combines disruption with extortion, often including data theft.
Practical use, context & sources
Reasonable steps to preserve ESI
ESI preservation
The practical preservation measures relevant to the Rule 37(e) inquiry when ESI that should have been preserved is lost and cannot be restored or replaced.
Practical use, context & sources
Reasonably foreseeable misuse
foreseeable misuse
Use outside an AI system's intended purpose that can reasonably follow from predictable human behavior or interaction with other systems.
Practical use, context & sources
Recognised legitimate interest (UK)
recognized legitimate interest
A separate UK processing basis introduced by DUAA for specified recognized purposes, subject to its statutory conditions.
Practical use, context & sources
Recommendation system
recommender
A system that ranks or suggests items, content or actions based on modeled relevance or preference. It may use supervised, collaborative, contextual or reinforcement methods.
Practical use, context & sources
Records of processing activities
RoPA
Documentation of processing purposes, data and individual categories, recipients, transfers, retention and safeguards under applicable recordkeeping duties.
Practical use, context & sources
Recovery
Restoring operations to an acceptable, validated state after disruption or compromise, including monitoring for recurrence.
Practical use, context & sources
Recovery point objective
RPO
The point in time to which data must be recovered, commonly expressed as a maximum acceptable data-loss interval.
Practical use, context & sources
Recovery time objective
RTO
The target elapsed time for restoring a system or capability after disruption, as defined by the recovery plan.
Practical use, context & sources
Rectification
right to correct
The right to have inaccurate personal information corrected and, where applicable, incomplete information completed.
Practical use, context & sources
Red teaming
Structured adversarial evaluation to find weaknesses under realistic threat assumptions. It differs from a normal acceptance test or a promise that all failures have been found.
Practical use, context & sources
Register of information (DORA)
RoI
The structured record of contractual arrangements for ICT services required for the relevant entity, sub-consolidated and consolidated levels.
Practical use, context & sources
Regulatory applicability
Determination that a particular legal duty applies to an entity, role, activity, jurisdiction or data use, including exceptions and transition rules.
Practical use, context & sources
Reinforcement learning
RL
Learning a policy for choosing actions from feedback about cumulative reward through interaction with an environment. A reward is a training signal, not a complete statement of what is lawful or desirable.
Practical use, context & sources
Reinforcement learning from AI feedback
RLAIF
Training approaches that use AI-generated evaluations or preferences as a feedback signal. They can scale evaluation but inherit assumptions and errors from the evaluating systems.
Practical use, context & sources
Reinforcement learning from human feedback
RLHF
A family of techniques using human preferences or evaluations to shape model behavior, often through a reward model and reinforcement learning.
Practical use, context & sources
Remote biometric identification
RBI
Identification without a person's active involvement, typically at a distance, by comparing biometric data with a reference database.
Practical use, context & sources
Reranking
Reordering retrieved candidates using another relevance model or scoring process before choosing context for generation.
Practical use, context & sources
Residual risk
Risk remaining after implemented controls or responses, subject to their actual effectiveness and uncertainty.
Practical use, context & sources
Resilience Imperative
Rich Davis's governance principle that an enterprise must be able to stop, reconstruct and recover from an agentic workflow, and that task persistence must remain subordinate to enterprise authority.
Practical use, context & sources
Retrieval-augmented generation
RAG
A system pattern that retrieves external information and supplies it as context to a generative model. Retrieval updates available knowledge without necessarily changing model weights.
Practical use, context & sources
Reward hacking
specification gaming
Achieving a high reward or apparent success by exploiting weaknesses in the objective or measurement instead of producing the intended outcome.
Practical use, context & sources
Right to correct
rectification request
An individual's ability under an applicable privacy law to request correction of inaccurate personal information, subject to its scope and verification rules.
Practical use, context & sources
Risk appetite
The types and amount of risk an organization is willing to pursue or retain in achieving objectives, expressed through its governance process.
Practical use, context & sources
Risk tolerance
Acceptable variation or limits around objectives and risk measures. It makes broad risk appetite more operational.
Practical use, context & sources
Robotic process automation
RPA
Automation of predefined application or user-interface tasks. RPA may be deterministic; adding an AI decision component changes the risk profile but does not make every RPA workflow agentic.
Practical use, context & sources
Role-based access control
RBAC
Permissions assigned to roles and then to users or workloads based on their responsibilities.
Practical use, context & sources
Safe Harbor de-identification (HIPAA)
HIPAA Safe Harbor
A de-identification route requiring removal of listed identifiers and no actual knowledge that remaining information can identify an individual.
Practical use, context & sources
Safe-state recovery
A process-owner-approved path from interrupted or unsafe operation to a validated condition in which essential activity can continue safely.
Practical use, context & sources
Safety evaluation
Testing whether a system behaves within defined harm-prevention criteria for specified contexts.
Practical use, context & sources
Sale of personal information (CCPA)
CCPA sale
A defined exchange or making available of personal information for monetary or other valuable consideration, subject to statutory exceptions.
Practical use, context & sources
Sandbox
An execution environment designed to constrain code, data access and effects. Isolation quality depends on configuration and the available escape routes.
Practical use, context & sources
SAR confidentiality
SAR nondisclosure
Restrictions protecting a suspicious activity report and information that would reveal its existence, subject to permitted disclosures.
Practical use, context & sources
Security information and event management
SIEM
A platform and process for collecting, normalizing, correlating, searching and retaining security events across sources.
Practical use, context & sources
Security operations center
SOC
The people, processes and systems that monitor security, investigate events and coordinate response.
Practical use, context & sources
Security orchestration, automation and response
SOAR
Integration and automation of security workflows, such as enrichment, triage, approval and containment.
Practical use, context & sources
Self-supervised learning
SSL
Learning in which supervisory signals are constructed from the input data itself, such as predicting hidden tokens or relationships between transformed views. It underpins many large-scale pretraining methods.
Practical use, context & sources
Semi-supervised learning
Learning that combines a relatively small labeled dataset with unlabeled examples. Pseudo-labeling is one approach, but inaccurate inferred labels can propagate errors.
Practical use, context & sources
Senior governing body (NYDFS)
senior governing body
The board, appropriate committee or equivalent governing body, with specified alternatives where no such body exists and for adopted affiliate programs.
Practical use, context & sources
Sensitive personal data
special-category data
Personal information subject to heightened protection under an applicable law. Categories differ; GDPR special categories include health, genetics and specified biometric or belief-related data.
Practical use, context & sources
Serious incident (EU AI Act)
AI serious incident
An AI-system incident or malfunction leading to specified serious consequences for health, critical infrastructure, fundamental-rights obligations, property or the environment.
Practical use, context & sources
Service provider (CCPA)
CCPA service provider
An entity processing personal information for a business under the CCPA's contract, purpose and use restrictions.
Practical use, context & sources
Severe but plausible scenario
severe-but-plausible testing
A challenging but credible disruption used to assess resilience against the relevant service impact tolerance.
Practical use, context & sources
Sharing (CCPA)
CCPA sharing
Making personal information available for cross-context behavioral advertising as defined by the CCPA, whether or not payment is involved.
Practical use, context & sources
Significant Data Fiduciary
SDF
A Data Fiduciary or class notified by India's Central Government under the statutory designation process.
Practical use, context & sources
SLH-DSA
A stateless hash-based digital signature algorithm standardized in FIPS 205, with different performance and security-design tradeoffs from ML-DSA.
Practical use, context & sources
Small language model
SLM
A relative industry label for a language model with a smaller parameter, memory or compute footprint than larger alternatives. There is no universal size cutoff or inherent safety advantage.
Practical use, context & sources
Software bill of materials
SBOM
A structured inventory describing software components and their relationships. It enables transparency but is not itself evidence that software is secure.
Practical use, context & sources
Software composition analysis
SCA
Analysis of third-party software components and dependencies for issues such as known vulnerabilities and licensing obligations.
Practical use, context & sources
Special categories of personal data
special-category data
The sensitive categories identified by GDPR Article 9, whose processing requires an applicable Article 9 condition as well as a lawful basis where required.
Practical use, context & sources
Specified purpose (India DPDP)
specified purpose
The purpose stated in the notice given by the Data Fiduciary in accordance with the DPDP Act and its rules.
Practical use, context & sources
Standard contractual clauses
SCCs
European Commission-approved contractual clauses that can support specified international transfers, subject to their conditions and accompanying assessment requirements.
Practical use, context & sources
Static application security testing
SAST
Analysis of source code or compiled artifacts for security weaknesses without executing the full application in its normal runtime environment.
Practical use, context & sources
Storage limitation
retention
Keeping identifiable personal data only as long as justified by purpose and applicable obligations, subject to specific lawful exceptions.
Practical use, context & sources
Subprocessor
A processor engaged by another processor to carry out processing for a controller.
Practical use, context & sources
Substantial modification (EU AI Act)
substantial change
An unplanned change after market placement or service entry that affects compliance with high-risk requirements or changes the assessed intended purpose.
Practical use, context & sources
Supervised learning
Learning from examples with target labels or outcomes, so a model predicts the target for new inputs. Classification and regression are common supervised tasks.
Practical use, context & sources
Support period (CRA)
security support period
The period during which the manufacturer is responsible for handling a product's vulnerabilities under the applicable CRA requirements.
Practical use, context & sources
Suspicious activity report
SAR
A regulated report to FinCEN about activity meeting the applicable institution-specific suspicion, transaction and reporting criteria.
Practical use, context & sources
Symbolic AI
rule-based AI / GOFAI
AI that represents knowledge using explicit symbols, logical statements or rules and applies reasoning procedures to derive conclusions. Its behavior can depend heavily on the completeness and consistency of the encoded knowledge.
Practical use, context & sources
Synthetic data
Artificially generated data intended to resemble some properties of reference data. It can preserve training artifacts or disclose information and is not automatically anonymous.
Practical use, context & sources
System instruction
system prompt
A higher-priority instruction layer used by an application to define behavior or constraints. Its effectiveness depends on model and system design, and it is not a secure authorization boundary.
Practical use, context & sources
System security plan
SSP
A document describing the system, its security requirements and the controls in place or planned to meet them.
Practical use, context & sources
Tactics, techniques and procedures
TTPs
Tactics describe adversary objectives, techniques how they are pursued, and procedures their concrete implementations.
Practical use, context & sources
Tamper-evident record
A record designed so unauthorized alteration can be detected. Tamper-evident is not the same as impossible to alter or impossible to delete.
Practical use, context & sources
Temperature
A sampling parameter that changes the distribution from which generated tokens are selected. Lower settings often reduce variation but do not guarantee factuality or determinism.
Practical use, context & sources
Third-party service provider (NYDFS)
TPSP
A non-affiliate, non-governmental service provider meeting Part 500's conditions, including access to nonpublic information through its services.
Practical use, context & sources
Threat
A circumstance or event with potential to adversely affect operations, assets or individuals through unauthorized access, disruption, disclosure or other harm.
Practical use, context & sources
Threat modeling
Structured analysis of assets, architecture, trust boundaries, threats and controls before and during system development.
Practical use, context & sources
Threat-led penetration testing
TLPT
A controlled assessment using threat intelligence to test relevant live production systems and functions under the prescribed DORA framework for selected entities.
Practical use, context & sources
Token
A unit used by a model's tokenizer, which may represent part of a word, a whole word, punctuation or another encoded element. Token counts are not the same as word counts.
Practical use, context & sources
Tool invocation
function calling
A structured request for an external function or service, followed by execution by the surrounding software. A model proposing a tool call is not itself an authorization decision.
Practical use, context & sources
Traceability
The ability to follow a decision, artifact or change through its lifecycle and relationships to sources and responsible actors.
Practical use, context & sources
Training
Fitting a model's parameters or decision structure using data, objectives and an optimization process. It is distinct from simply sending a request to an already deployed model.
Practical use, context & sources
Trajectory conformance
Assessment of the sequence of actions against an approved operating mandate, not solely assessment of the final answer.
Practical use, context & sources
Transfer learning
Reusing knowledge or representations learned for one task or dataset in another task, often by adapting a pretrained model.
Practical use, context & sources
Transformer
A neural architecture built around attention mechanisms for representing relationships within data sequences. Transformers can support language, vision and other tasks; the architecture itself does not confer agency.
Practical use, context & sources
Transparency
Making appropriate information about AI purpose, operation, limitations and governance available to relevant parties.
Practical use, context & sources
Trust boundary
A point at which identity, authority or trust assumptions change between systems, components or organizations.
Practical use, context & sources
Unsafe output handling
Passing AI output into software, databases, browsers or commands without validation appropriate to the receiving context.
Practical use, context & sources
Unsupervised learning
Learning structure from data without an externally supplied target label for each example. Clustering, dimensionality reduction and some anomaly-detection methods fit this category.
Practical use, context & sources
User and entity behavior analytics
UEBA
Analysis of behavioral patterns for users, workloads or other entities to identify anomalies or risks.
Practical use, context & sources
Validation
Assessing whether a system, model or process is suitable for its intended purpose in the relevant context.
Practical use, context & sources
Vector database
vector store
A data system that indexes vectors and often associated text and metadata for similarity retrieval. It may include approximate nearest-neighbor methods.
Practical use, context & sources
Verification
Checking whether specified requirements have been met, often by inspection or testing.
Practical use, context & sources
Vital interests
vital-interest basis
A lawful-processing ground associated with protecting a person's life or essential interests, within the law's conditions.
Practical use, context & sources
Vulnerability
A weakness in a system, process or control that a threat could exploit or that could lead to adverse outcomes.
Practical use, context & sources
Vulnerability handling
vulnerability management process
The organized identification, reporting, remediation and disclosure of product vulnerabilities throughout the relevant support lifecycle.
Practical use, context & sources
Work-product protection
work product
Protection for qualifying litigation-preparation materials under the relevant doctrine or rule, with different treatment for factual and mental-impression material.
Practical use, context & sources
Working memory
session state
Information retained for continuity within a workflow or conversation. It differs from model weights and may include sensitive inputs, tool results and intermediate decisions.
Practical use, context & sources
Write once, read many
WORM
A storage behavior that prevents modification or deletion under specified retention controls. Protection depends on the mode and administrative configuration.
Practical use, context & sources
Written information security program
WISP
A documented program of administrative, technical and physical safeguards designed for the institution's risks and the applicable rule's requirements.
Practical use, context & sources
Zero trust
ZT / ZTA
An approach that grants no implicit trust solely because of network location or ownership and evaluates access to resources using identity and context.
Practical use, context & sources
Zero-day vulnerability
A vulnerability not previously known to relevant defenders or without an available fix at the time of exposure or exploitation. Usage varies by context.
Practical use, context & sources
OpenTelemetry
OTel
A vendor-neutral observability project providing APIs, SDKs, protocols and tools for producing and transporting telemetry. It is not a SIEM, access-control system or assurance certification.
Practical use, context & sources
Span
A timed record of an operation, carrying context, attributes, status and potentially events or links. A span represents the operation its instrumentation observed.
Practical use, context & sources
Trace ID
trace_id / TraceId
An identifier correlating spans in a distributed trace. It joins telemetry; it is not an authentication credential, authorization token or cryptographic proof of origin.
Practical use, context & sources
Span ID
span_id / SpanId
The identifier for an individual span within its trace. Parent references and links describe relationships between observed operations.
Practical use, context & sources
Span context
Trace identity, span identity and associated tracing state that can be propagated between execution boundaries.
Practical use, context & sources
Parent and child spans
parent span / child span
A hierarchy describing how instrumented operations are nested or caused. It does not necessarily include every action in a business transaction.
Practical use, context & sources
Span link
linked traces
An explicit relationship from one span to another span context, including operations in a different trace.
Practical use, context & sources
W3C Trace Context
traceparent / tracestate
A standard format for propagating tracing information between services, including traceparent and tracestate headers.
Practical use, context & sources
Baggage
Contextual key-value data that can travel with distributed requests. It is separate from span attributes and may cross organizational boundaries.
Practical use, context & sources
Semantic conventions
semconv
Shared names, types and meanings for telemetry. Conventions describe a reporting vocabulary, not proof that controls exist or instrumentation is complete.
Practical use, context & sources
Development stability status
Development status
A maturity designation for technical conventions that can change before stabilization. It is not a statement about the maturity of the overall OpenTelemetry project.
Practical use, context & sources
Required attribute
A field an instrumentation must populate for a specified convention under its documented rules. Requirement levels apply to the relevant operation, not to every span universally.
Practical use, context & sources
Conditionally required attribute
A field required when the condition specified for that operation is met, such as an error occurring or a value being available.
Practical use, context & sources
Recommended attribute
A field the convention recommends collecting under its defined conditions; implementation support and configuration still determine availability.
Practical use, context & sources
Opt-in content capture
content capture / OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT
Explicitly enabled collection of potentially sensitive input, output, tool, retrieval or memory content where the applicable instrumentation supports it.
Practical use, context & sources
Resource attributes
service.name / service.version / service.instance.id / deployment.environment.name
Metadata describing the entity that produced telemetry, such as a service, instance, environment or container.
Practical use, context & sources
Instrumentation scope
The identity and version of the instrumentation library or component creating telemetry.
Practical use, context & sources
OpenTelemetry Protocol
OTLP / OTLP/HTTP / OTLP/gRPC
The protocol used to exchange telemetry between producers, Collectors and backends, commonly over HTTP or gRPC.
Practical use, context & sources
OpenTelemetry Collector
collector
A configurable service that receives, processes and exports telemetry to one or more destinations.
Practical use, context & sources
Telemetry receiver
OTel receiver
A Collector component accepting incoming telemetry using a supported protocol or source.
Practical use, context & sources
Telemetry processor
OTel processor
A pipeline component that transforms, filters, batches or otherwise processes received telemetry.
Practical use, context & sources
Telemetry exporter
OTel exporter
A component sending telemetry to a destination backend.
Practical use, context & sources
Telemetry connector
spanmetrics connector
A Collector component that connects pipelines, potentially deriving one signal from another.
Practical use, context & sources
Head sampling
A sampling decision made early in a trace using information available at that point.
Practical use, context & sources
Tail sampling
tail-based sampling
A sampling approach that uses accumulated trace information before deciding what to retain.
Practical use, context & sources
Evidence-retention contract
An organization-defined statement of which actions and records must be retained, by which producers, for which purpose and period.
Practical use, context & sources
Collector internal telemetry
Measurements and logs about the telemetry pipeline itself, rather than about the applications it observes.
Practical use, context & sources
Exporter queue
sending queue / persistent queue
A bounded buffer holding telemetry awaiting delivery. Queues and retries reduce some loss conditions but do not guarantee delivery.
Practical use, context & sources
Telemetry backpressure
backpressure
A condition in which incoming telemetry exceeds downstream processing or delivery capacity.
Practical use, context & sources
Telemetry redaction
redaction
Removing or replacing sensitive values before they reach an unauthorized destination or audience.
Practical use, context & sources
Hashing and pseudonymization
hashed metadata
Hashing can support matching or change detection without displaying the original value, but predictable inputs can remain identifiable or guessable.
Practical use, context & sources
Content-addressed identifier
artifact digest / container.image.digest
An identifier derived from an artifact digest, allowing comparisons against the same bytes.
Practical use, context & sources
Independent producer corroboration
Comparing an application report with records generated by separately controlled gateways, brokers or downstream services.
Practical use, context & sources
Model snapshot
pinned model
A versioned model reference or artifact identifying the model configuration intended or observed in a deployment.
Practical use, context & sources
Requested versus responding model
gen_ai.request.model / gen_ai.response.model
The distinction between the model identified in a request and the model identified in the observed response.
Practical use, context & sources
Provider routing identity
gen_ai.provider.name / server.address / server.port
The provider and endpoint reported by instrumentation, which may identify a gateway rather than the ultimate model provider.
Practical use, context & sources
GenAI operation name
gen_ai.operation.name
The operation class reported in gen_ai.operation.name, such as inference, retrieval or tool execution under a selected convention.
Practical use, context & sources
Tool-call identity
gen_ai.tool.call.id / gen_ai.tool.name
An identifier joining a model tool request and its execution record where the instrumentation and application preserve it.
Practical use, context & sources
GenAI response identity
gen_ai.response.id
A provider- or service-assigned identifier for a particular response, where available.
Practical use, context & sources
Finish reason
gen_ai.response.finish_reasons
An instrumented indication of how generation ended, interpreted in the relevant provider and convention context.
Practical use, context & sources
Token usage
gen_ai.usage.input_tokens / gen_ai.usage.output_tokens
Counts reported for model input and output, with provider-specific accounting details such as cached tokens.
Practical use, context & sources
Error type
error.type
A low-cardinality classification of an operation error reported by instrumentation.
Practical use, context & sources
Retrieval source identity
gen_ai.data_source.id / gen_ai.retrieval.top_k
The identifier of a grounding source reported by a retrieval operation.
Practical use, context & sources
Memory operation
search_memory / create_memory / update_memory / upsert_memory / delete_memory / delete_memory_store
An observed read, creation, update, upsert or deletion involving an agent memory record or store.
Practical use, context & sources
GenAI evaluation result event
gen_ai.evaluation.result / gen_ai.evaluation.name / gen_ai.evaluation.score.value
A structured result evaluating an output for a named quality or safety characteristic. The GenAI convention includes gen_ai.evaluation.result.
Practical use, context & sources
Custom control span
practical_trust.*
An application-defined span for a control operation not covered by the selected standard convention.
Practical use, context & sources
Authorization receipt
decision.id / policy.digest
A proposed record of exactly what the system was allowed to do: who requested it, the permitted action and destination, the approved inputs, the rule used and when permission expires. The system carrying out the action checks this record before acting.
Practical use, context & sources
Delegation envelope
authority envelope
A proposed bounded grant carrying permitted scope, duration, spending or execution limits and revocation conditions to a delegated agent.
Practical use, context & sources
Evaluation suite version
The identifier of the tests, datasets, rubric and evaluator configuration used to assess a release or observed output.
Practical use, context & sources
Quality drift
behavioral drift
A material change in useful or safe output behavior over time, even when provider and model metadata appear unchanged.
Practical use, context & sources
Configuration drift
A difference between approved configuration and observed versions, routes, tool sets, policies or other deployment settings.
Practical use, context & sources
Canary evaluation
canary traffic
A controlled evaluation or limited release exposure used to look for regressions before expanding deployment.
Practical use, context & sources
Authority and consumption budget
denial-of-wallet / spend cap
A proposed enforceable limit on the resources and delegated spending available to an agent task or task family.
Practical use, context & sources
Retrieval index manifest
index.manifest.hash
A versioned inventory of the corpus, chunking, embedding configuration, permissions and index artifacts used for retrieval.
Practical use, context & sources
Agent retirement evidence
deprovisioning
Records showing that an agent deployment and its residual grants, schedules, credentials and retained state were closed or reassigned under policy.
Practical use, context & sources
Transparency and recourse record
A proposed decision-linked record supporting an understandable explanation and an appropriate route for review or challenge.
Practical use, context & sources
Telemetry coverage
The observed extent to which expected services and actions produce usable evidence. Missing telemetry can reflect absent activity, instrumentation gaps or pipeline loss.
Practical use, context & sources
Span-derived metrics
span-to-metrics
Aggregated measurements computed from spans, such as request counts or duration distributions.
Practical use, context & sources
Metric cardinality
high cardinality
The number of distinct time series produced by combinations of metric attribute values.
Practical use, context & sources
Baseline percentile
p99 / rolling baseline
A statistical threshold derived from a defined reference population and period, such as the 99th percentile of output volume.
Practical use, context & sources
Derived security event
A curated security record produced by evaluating operational evidence against a rule, model or control condition.
Practical use, context & sources
Open Cybersecurity Schema Framework
OCSF
An open schema for security-event representation. An OpenTelemetry signal requires an explicit mapping before it can be represented as an OCSF event.
Practical use, context & sources
Elastic Common Schema
ECS
A field convention for structuring events in Elastic-oriented analysis. It remains a separate schema whose fields can be mapped to other telemetry conventions.
Practical use, context & sources
Logical-operation retry
retry visibility
An automatic retry within an instrumented logical operation. A GenAI operation span can cover the complete operation rather than one span per attempt.
Practical use, context & sources
Tiered telemetry storage
A proposed separation of metrics, metadata traces, security events and restricted content according to purpose, sensitivity and retention needs.
Practical use, context & sources
Telemetry policy floor
An organization-defined minimum evidence and privacy contract enforced through instrumentation, gateways and approved pipeline configuration.
Practical use, context & sources
THE PRACTICAL TRUST REFERENCE DESK
AI & technology regulatory library
Find the questions your organization needs to resolve before it accepts technology risk. Start with the business activity, then examine jurisdiction, role, obligations and the evidence needed to support oversight.
131 regulatory and framework records. Preserved source-review dates are displayed on individual records; this redesign is not a new legal audit. Sector and NAICS filters are research aids, not legal applicability determinations.
Choose your sector
Scroll to browse 131 regulatory and framework records.
FTC Act, Section 5
Businesses within Federal Trade Commission jurisdiction. Sectoral exemptions and the distinction between commercial conduct and regulated banking or nonprofit activity matter.
Status: In force; case-specific enforcement
Duties, evidence, limitations & official sources
California Consumer Privacy Act / CPRA
For-profit businesses meeting statutory California thresholds; service-provider and contractor duties differ. Exemptions are often data-specific, not blanket exemptions for an entire enterprise.
Status: CCPA/CPRA in force; 2025 final regulations effective January 1, 2026 with later compliance/submission dates
Duties, evidence, limitations & official sources
New York SHIELD Act and breach notification
Persons and businesses holding covered private information of New York residents; reasonable-security duties and notification triggers should be assessed separately.
Status: In force
Duties, evidence, limitations & official sources
Virginia Consumer Data Protection Act
Controllers meeting Virginia resident-data or revenue thresholds, subject to statutory entity and data exemptions.
Status: In force
Duties, evidence, limitations & official sources
Colorado Privacy Act
Controllers meeting statutory coverage tests with Colorado nexus; not every small business or employment record falls in scope.
Status: In force; distinct from Colorado AI legislation
Duties, evidence, limitations & official sources
Connecticut Data Privacy Act
Covered businesses processing Connecticut resident personal data; coverage, exempt entities and processing purpose must be evaluated under amended CTDPA and specific new provisions.
Status: In force; Public Act 26-64 (SB4) introduced additional requirements beginning October 1, 2026, with provision-specific timing
Duties, evidence, limitations & official sources
Texas Data Privacy and Security Act
Persons conducting business in Texas or producing products/services consumed by residents, subject to small-business and other statutory exclusions and exceptions.
Status: In force; separate from Texas AI law
Duties, evidence, limitations & official sources
Washington My Health My Data Act
Covered consumer-health-data businesses and processors, including some apps and inferences outside HIPAA. The Act has its own nexus and exemption rules.
Status: In force
Duties, evidence, limitations & official sources
Illinois Biometric Information Privacy Act
Private entities collecting or holding covered biometric identifiers/information, with statutory exclusions. Face recognition and voiceprint applications need classification beyond ordinary photos/audio.
Status: In force; amended liability provisions require current-text review
Duties, evidence, limitations & official sources
Children's Online Privacy Protection Act and Rule
Operators of child-directed online services and operators with actual knowledge of collecting personal information from children under 13.
Status: In force; use current amended rule
Duties, evidence, limitations & official sources
Family Educational Rights and Privacy Act
Covered educational agencies and institutions receiving applicable US Department of Education funding; vendors need a valid disclosure basis, often the school-official exception.
Status: In force
Duties, evidence, limitations & official sources
Title VII and employment discrimination
Covered employers, employment agencies and labor organizations. AI recruitment, assessment, promotion and termination tools remain employment-selection practices.
Status: In force
Duties, evidence, limitations & official sources
Americans with Disabilities Act
Covered employers, public entities and public accommodations under the relevant ADA title. Disability-sensitive algorithmic screening and digital accessibility need separate assessments.
Status: In force
Duties, evidence, limitations & official sources
New York City Local Law 144 - automated employment decisions
Employers and employment agencies using an AEDT within the law and rules' definition and geographic scope; not every workplace AI tool qualifies.
Status: Enforced since July 5, 2023
Duties, evidence, limitations & official sources
Texas Responsible Artificial Intelligence Governance Act
Entities and AI systems within Texas jurisdiction as defined; government, healthcare and other provisions have differing scopes. Do not confuse this with TDPSA.
Status: Effective January 1, 2026
Duties, evidence, limitations & official sources
Colorado automated decisionmaking law - SB26-189
Covered developers and deployers using automated decision-making technology that materially influences consequential decisions in employment, education, housing, lending, insurance, health care, and government services; exemptions must be checked.
Status: Enacted May 14, 2026; principal covered-ADMT duties begin January 1, 2027; AG implementing rules proposed as of October 27
Duties, evidence, limitations & official sources
California Transparency in Frontier Artificial Intelligence Act
Frontier developers and large frontier developers meeting statutory model/developer thresholds. Not a general certification requirement for enterprise AI users.
Status: In force; frontier-developer and large-frontier-developer classifications control
Duties, evidence, limitations & official sources
California AI Transparency Act — SB 1000 amendment (2026)
Covered generative-AI providers and selected licensees/platform roles under California Business and Professions Code §§ 22757.1–22757.5, as amended. Do not rely on obsolete user-count thresholds.
Status: SB1000 Chapter 861 enacted and effective September 30, 2026; current operative duties depend on amended section and covered actor
Duties, evidence, limitations & official sources
New York Responsible AI Safety and Education (RAISE) Act
Frontier and large frontier developers within Article 44-B definitions and scope; applicability is not equivalent to deploying any enterprise LLM.
Status: Enacted; operative January 1, 2027
Duties, evidence, limitations & official sources
NYDFS Cybersecurity Regulation
Entities operating under covered NY financial-services laws; limited exemptions and Class A requirements require a documented scope determination.
Status: In force; amended requirements and exemptions apply
Duties, evidence, limitations & official sources
GLBA and FTC Safeguards Rule
Financial institutions subject to the FTC's GLBA jurisdiction, including covered nonbank financial activities. Banking-agency rules govern other covered institutions.
Status: In force
Duties, evidence, limitations & official sources
SEC Regulation S-P
Covered broker-dealers, funding portals, investment companies, SEC-registered investment advisers and covered transfer agents. Size classifications differ by registrant type; the adviser AUM threshold is not a universal threshold for all entities.
Status: 2024 amendments: larger entities December 3, 2025; smaller entities June 3, 2026; both dates elapsed
Duties, evidence, limitations & official sources
SEC cybersecurity disclosure rules
Public-company registrants subject to the relevant Exchange Act reporting forms; domestic and foreign private issuer mechanisms differ.
Status: In force for covered registrants
Duties, evidence, limitations & official sources
Interagency model risk management guidance - SR26-2
Interagency guidance most relevant to supervised banking organizations with more than $30 billion in total assets; smaller organizations with significant model exposure should consider applicability. The guidance covers complex quantitative models, including non-generative, non-agentic AI models. Footnote 3 expressly excludes generative and agentic AI from its scope.
Status: Interagency guidance issued April 17, 2026; replaces SR11-7 and SR21-8; not a new agentic-AI rule
Duties, evidence, limitations & official sources
Interagency third-party relationship risk management
Banking organizations supervised by the Federal Reserve, FDIC or OCC; lifecycle practices scaled to the risk of each relationship.
Status: Final interagency guidance, 2023
Duties, evidence, limitations & official sources
Equal Credit Opportunity Act / Regulation B
Covered creditors making credit decisions, including AI-enabled underwriting. Technology choice does not eliminate statutory notification duties.
Status: In force; current regulation controls
Duties, evidence, limitations & official sources
Fair Credit Reporting Act
Consumer reporting agencies, furnishers and users of consumer reports as defined; AI scoring vendors can create reporting-law obligations depending on their actual functions.
Status: In force
Duties, evidence, limitations & official sources
Bank Secrecy Act / AML program requirements
Covered financial institutions, including covered MSBs. AML program, customer due-diligence and reporting obligations differ by institution type.
Status: In force; institution-specific rules apply
Duties, evidence, limitations & official sources
US economic sanctions / OFAC compliance
US persons and transactions subject to applicable sanctions jurisdiction; extraterritorial and secondary-sanctions issues are fact-specific.
Status: In force; program-specific rules change frequently
Duties, evidence, limitations & official sources
NAIC model bulletin on insurer use of AI
Insurers in states adopting the bulletin or related guidance. NAIC publication alone is not a nationwide enforceable statute.
Status: Model bulletin; state adoption and variations control
Duties, evidence, limitations & official sources
PCI DSS v4.0.1
Entities storing, processing or transmitting payment-card account data, and entities able to affect its security, as scoped by payment-brand/acquirer obligations.
Status: Contractual standard; not a general statute
Duties, evidence, limitations & official sources
HIPAA Privacy and Breach Notification Rules
Covered entities and business associates handling protected health information. Not every wellness app or AI-generated health inference is HIPAA-covered.
Status: In force; scope and exceptions are specific
Duties, evidence, limitations & official sources
HIPAA Security Rule
Covered entities and business associates safeguarding electronic PHI. Security-rule obligations are separate from general medical-device and consumer-app regulation.
Status: Existing rule in force; do not treat the 2025 NPRM as final law
Duties, evidence, limitations & official sources
FTC Health Breach Notification Rule
Vendors of personal health records, related entities and service providers covered by the rule, often outside HIPAA.
Status: In force; amended coverage requires specific analysis
Duties, evidence, limitations & official sources
Medical-device cybersecurity - FDCA Section 524B
Manufacturers submitting covered cyber devices under the FDCA. Scope depends on the device and premarket pathway, not whether the product simply uses AI.
Status: In force for applicable cyber-device submissions
Duties, evidence, limitations & official sources
FDA electronic records and electronic signatures
Electronic records/signatures within FDA Part 11 scope. Not every document at a regulated company is automatically a Part 11 record.
Status: In force; predicate-rule applicability matters
Duties, evidence, limitations & official sources
FDA Quality Management System Regulation
Covered finished medical-device manufacturers. Current Part 820 incorporates ISO 13485:2016 with FDA-specific requirements.
Status: QMSR effective February 2, 2026; current 21 CFR Part 820
Duties, evidence, limitations & official sources
FDA drug CGMP and laboratory records
Drug manufacturing within 21 CFR Parts 210/211; laboratory and computerized-system obligations depend on the regulated activity and predicate rules.
Status: In force
Duties, evidence, limitations & official sources
NERC Critical Infrastructure Protection standards
Registered entities and applicable Bulk Electric System cyber assets. These requirements do not automatically cover every local utility or industrial-control asset.
Status: Applicable enforceable versions depend on jurisdiction and registration
Duties, evidence, limitations & official sources
Water-system risk and resilience / emergency planning
Community drinking-water systems serving more than 3,300 people under SDWA Section 1433; smaller systems receive guidance but are not subject to the same certification threshold.
Status: In force; five-year certification cycle
Duties, evidence, limitations & official sources
Cyber Incident Reporting for Critical Infrastructure Act
Entities and incidents ultimately covered by CISA's implementing regulations. Consult the live rulemaking record before assigning a compliance start date.
Status: Implementing final rule under OIRA review; no operative final-rule reporting commencement verified
Duties, evidence, limitations & official sources
Federal Information Security Modernization Act
Federal information and systems, including covered contractor-operated systems. National-security systems have distinct requirements.
Status: In force
Duties, evidence, limitations & official sources
OMB M-25-21 - federal AI use
Covered executive agencies and AI uses under the memorandum, including its exclusions; not a private-sector AI statute.
Status: Issued April 3, 2025; agency-specific implementation applies
Duties, evidence, limitations & official sources
OMB M-25-22 - federal AI acquisition
Covered federal AI acquisitions and agency procurement processes. Contractors are affected through applicable procurement terms.
Status: Issued April 3, 2025; replaces M-24-18
Duties, evidence, limitations & official sources
DFARS covered defense information and incident reporting
Contractors/subcontractors subject to the relevant DFARS clause and covered defense information or operationally critical support.
Status: Applies when included in the contract
Duties, evidence, limitations & official sources
Cybersecurity Maturity Model Certification program
Defense contractors/subcontractors when solicitation/contract requirements specify a CMMC level; related program and acquisition rules work together.
Status: Phase II implementation suspended July 13, 2026; Phase I self-assessment requirements remain
Duties, evidence, limitations & official sources
Privacy Act of 1974 and E-Government privacy assessments
Federal agency systems of records and covered information technology processing; contractor roles and specific exemptions require analysis.
Status: In force
Duties, evidence, limitations & official sources
Professional conduct - competence, confidentiality and supervision
Lawyers and law firms under the rules of their admitting/practicing jurisdictions. ABA Model Rules are not automatically enacted nationwide.
Status: Binding rules are jurisdiction-specific; ABA text is a model
Duties, evidence, limitations & official sources
Federal civil procedure - ESI and preservation
Parties and counsel in federal civil litigation; preservation duties depend on foreseeability, control and the circumstances.
Status: In force; case orders and current rules control
Duties, evidence, limitations & official sources
Delaware fiduciary oversight jurisprudence
Directors/officers and entities governed by relevant Delaware law. Applicable duties, exculpation and pleading standards require case-specific advice.
Status: Fact-specific; not automatic liability after an incident
Duties, evidence, limitations & official sources
European Union Artificial Intelligence Act
Providers, deployers, importers, distributors and other covered actors; certain non-EU actors are included. System role, prohibited use, high-risk classification and GPAI status determine duties.
Status: In force; current amended application dates and legacy-system transitions differ by provision
Duties, evidence, limitations & official sources
General Data Protection Regulation
Controllers/processors within Article 3 territorial scope; non-EU establishments can be covered by offering goods/services or monitoring people in the Union.
Status: Applicable since May 25, 2018
Duties, evidence, limitations & official sources
ePrivacy Directive and national implementing laws
Electronic communications and related storage/access or direct-marketing activities under applicable member-state legislation.
Status: National implementation controls
Duties, evidence, limitations & official sources
Digital Operational Resilience Act
Covered financial entities and designated critical ICT third-party providers, subject to the Act's proportionality/exemptions and implementing measures.
Status: Applicable since January 17, 2025
Duties, evidence, limitations & official sources
NIS2 Directive
Essential/important entities across listed sectors, size rules and specific exceptions. A group may have different national registrations and competent authorities.
Status: EU directive in force; national transposition and scope control
Duties, evidence, limitations & official sources
Cyber Resilience Act
Covered products with digital elements placed on the EU market and relevant manufacturers/importers/distributors; exclusions and open-source distinctions matter.
Status: Reporting obligations applicable since September 11, 2026; main product requirements December 11, 2027
Duties, evidence, limitations & official sources
Digital Services Act
Intermediary services offered to recipients in the Union; hosting, platforms and very large services have different obligations.
Status: Applicable; duties depend on service category/designation
Duties, evidence, limitations & official sources
EU Data Act
Connected products/related services, data holders/users and data-processing service providers within the Act's scope.
Status: Generally applicable from September 12, 2025; provisions phased
Duties, evidence, limitations & official sources
EU Copyright in the Digital Single Market Directive
Rights holders, text/data-mining users and relevant platforms. Research and general text/data-mining exceptions have distinct conditions.
Status: National implementing law applies
Duties, evidence, limitations & official sources
Revised EU Product Liability Directive
Products, including software, within the revised regime once implemented and temporally applicable. AI-provider liability is not automatic for every output.
Status: Directive (EU) 2024/2853 in force; Member State transposition due December 9, 2026
Duties, evidence, limitations & official sources
EU Medical Device Regulation
Medical devices and medical-device software classified by intended purpose; an AI component does not by itself establish or remove device status.
Status: Applicable; device transition provisions vary
Duties, evidence, limitations & official sources
EU In Vitro Diagnostic Medical Device Regulation
IVD devices and relevant diagnostic software, with classification and legacy-device transitions determined under current law.
Status: Applicable; class-specific transition conditions vary
Duties, evidence, limitations & official sources
EU GMP Annex 11 - computerized systems
Computerized systems used in GMP-regulated activities under applicable medicines law and manufacturing authorization.
Status: Published Annex 11 Computerised Systems: January 2011 revision in the current EudraLex listing
Duties, evidence, limitations & official sources
EU Machinery Regulation
Machinery and related products within the EU product-safety framework; safety-related software and substantial modifications require scope review.
Status: Generally applies January 20, 2027; existing regime until transition
Duties, evidence, limitations & official sources
Critical Entities Resilience Directive
Critical entities designated by member states in specified sectors; complementary to NIS2 rather than interchangeable with it.
Status: National transposition/designation determines obligations
Duties, evidence, limitations & official sources
UK GDPR / Data Protection Act 2018 / DUAA 2025
Controllers/processors within UK territorial scope. The UK regime is not identical to the EU GDPR after domestic amendments.
Status: UK GDPR and DPA in force; all DUAA data-protection provisions in force by June 19, 2026
Duties, evidence, limitations & official sources
Privacy and Electronic Communications Regulations
Electronic marketing and covered communications/terminal-equipment activities. Apply current exemptions and DUAA-related changes.
Status: In force as amended
Duties, evidence, limitations & official sources
FCA operational resilience rules
FCA firms within SYSC 15A scope and corresponding prudential requirements where applicable.
Status: In force; March 31, 2025 transition completed for covered firms
Duties, evidence, limitations & official sources
FCA Consumer Duty
Firms and products/services within the FCA's retail-market scope, including relevant distribution-chain roles.
Status: In force for covered retail business
Duties, evidence, limitations & official sources
UK Online Safety Act
Covered user-to-user/search services with UK links, including relevant AI-powered service functions. Not all standalone AI enterprise software is in scope.
Status: Phased duties; service category and commencement matter
Duties, evidence, limitations & official sources
PIPEDA
Commercial personal-information processing within federal scope and cross-border activity; exemptions and provincial allocation require review.
Status: In force; substantially similar provincial regimes may apply
Duties, evidence, limitations & official sources
Quebec private-sector privacy law (Law 25 reforms)
Enterprises holding personal information within Quebec private-sector law; separate public-sector statute applies to public bodies.
Status: Reforms phased through 2024; current consolidated law controls
Duties, evidence, limitations & official sources
Canadian Directive on Automated Decision-Making
Federal departments and administrative decisions within the directive's scope; excludes or treats some uses differently. Not a private-sector national AI Act.
Status: Policy applies to covered federal automated decisions
Duties, evidence, limitations & official sources
OSFI B-13 - technology and cyber risk
Federally regulated financial institutions and relevant foreign branches, applied proportionately to their Canadian operations.
Status: Current supervisory guideline
Duties, evidence, limitations & official sources
Australian Privacy Act and Australian Privacy Principles
APP entities including covered agencies/businesses; turnover and sector exceptions matter. Not every small business is exempt.
Status: In force; new ADM privacy-policy duties commence December 10, 2026
Duties, evidence, limitations & official sources
APRA CPS 234 - Information Security
APRA-regulated entities, with responsibilities for information assets managed by related parties and third parties.
Status: Mandatory for covered APRA entities
Duties, evidence, limitations & official sources
APRA CPS 230 - Operational Risk Management
Covered APRA-regulated entities; current standard and service-provider transition arrangements apply.
Status: Current version in force July 1, 2026
Duties, evidence, limitations & official sources
Singapore Personal Data Protection Act
Organizations processing personal data within PDPA scope; public agencies have a different governance regime.
Status: In force
Duties, evidence, limitations & official sources
Singapore PDPC guidance on AI use of personal data
Organizations developing/deploying AI systems using personal data, including recommendations, decisions and generative uses where PDPA applies.
Status: Guidance interprets existing PDPA duties; not a separate AI statute
Duties, evidence, limitations & official sources
MAS Technology Risk Management Guidelines
Financial institutions within MAS supervision, with institution-specific cyber hygiene and technology notices where applicable.
Status: January 2021 Technology Risk Management Guidelines; current official publication page accessible
Duties, evidence, limitations & official sources
Japan Act on the Protection of Personal Information
Business operators and other entities within APPI scope; overseas handling and cross-border transfers have specific requirements.
Status: In force as amended; Japanese original controls
Duties, evidence, limitations & official sources
China Personal Information Protection Law
Personal-information handlers within territorial/extraterritorial statutory scope. Localization and transfer obligations depend on thresholds and current implementing measures.
Status: In force; official Chinese text controls
Duties, evidence, limitations & official sources
China Data Security Law
Data activities within statutory scope, with important/core-data and national-security implications beyond personal information.
Status: In force; classification/export rules require current local review
Duties, evidence, limitations & official sources
China Interim Measures for Generative AI Services
Generative AI services provided to the public in China; the measures exclude certain nonpublic internal research/application uses from their specific scope.
Status: Effective August 15, 2023
Duties, evidence, limitations & official sources
India Digital Personal Data Protection Act and Rules
Digital personal-data processing within the Act's territorial/extraterritorial scope. Confirm which statutory/rule provisions have commenced before treating all duties as operative.
Status: Rules notified November 2025; phased 18-month implementation
Duties, evidence, limitations & official sources
India CERT-In cybersecurity directions
Covered service providers, intermediaries, data centers, bodies corporate and government organizations; additional duties apply to specified services.
Status: 2022 directions; applicable scope and exemptions matter
Duties, evidence, limitations & official sources
Brazil General Personal Data Protection Law
Personal-data processing within LGPD nexus; different lawful bases and public/private-sector rules may apply.
Status: In force; ANPD regulations supplement the law
Duties, evidence, limitations & official sources
South Africa Protection of Personal Information Act
Responsible parties and operators within POPIA scope; special personal information and child data have additional constraints.
Status: In force
Duties, evidence, limitations & official sources
Swiss Federal Act on Data Protection
Private persons and federal bodies within FADP scope; cantonal public-sector regimes can differ.
Status: Revised Act effective September 1, 2023
Duties, evidence, limitations & official sources
New Zealand Privacy Act
Agencies and organizations within statutory scope, including relevant overseas agencies carrying on business in New Zealand.
Status: Privacy Act 2020 in force; IPP3A indirect-collection notice commenced May 1, 2026
Duties, evidence, limitations & official sources
NIST AI Risk Management Framework 1.0
Organizations designing, developing, deploying or using AI; voluntary unless incorporated into law, procurement or contract.
Status: AI RMF 1.0 remains the published final framework; revision work is ongoing
Duties, evidence, limitations & official sources
NIST Generative AI Profile
Organizations using generative models and applications; companion profile to the AI RMF.
Status: NIST AI600-1 final, July 26, 2024; voluntary Generative AI Profile
Duties, evidence, limitations & official sources
NIST Cybersecurity Framework
Organizations managing cyber risk, including business processes supported by AI.
Status: Published guidance or standard; binding only where adopted
Duties, evidence, limitations & official sources
NIST Privacy Framework - final 1.0 / draft 1.1
Organizations managing privacy risk across data-processing ecosystems.
Status: Version 1.0 final; version 1.1 Initial Public Draft remains draft in the reviewed publication record
Duties, evidence, limitations & official sources
NIST SP800-53 Rev.5 - security and privacy controls
Information systems within adopted security/privacy baselines; federal use and nonfederal contractual uses differ.
Status: Revision 5 final; minor Release 5.2.0 issued August 27, 2025
Duties, evidence, limitations & official sources
NIST SSDF - SP800-218 v1.1 / v1.2 draft
Software producers and acquirers, including AI harnesses, toolchains and dependencies.
Status: SSDF v1.1 final (2022); SP800-218 Rev. 1 / SSDF v1.2 Initial Public Draft (December 2025)
Duties, evidence, limitations & official sources
NIST incident response recommendations
Organizations integrating cyber incident response into risk management.
Status: SP800-61 Rev.3 final, April 2025; a CSF 2.0 Community Profile
Duties, evidence, limitations & official sources
NIST operational technology security guide
Industrial control, operational technology and cyber-physical environments.
Status: Published guidance or standard; binding only where adopted
Duties, evidence, limitations & official sources
NIST post-quantum cryptography standards
Cryptographic implementations and organizations with mandated or risk-based migration needs.
Status: Published guidance or standard; binding only where adopted
Duties, evidence, limitations & official sources
ISO/IEC 42001 AI management systems
Organizations adopting an AI management system, optionally subject to independent certification.
Status: ISO/IEC 42001:2023; voluntary management-system standard unless incorporated
Duties, evidence, limitations & official sources
ISO/IEC 23894 AI risk management
Organizations managing risks associated with AI products, systems and services.
Status: ISO/IEC 23894:2023 risk-management guidance
Duties, evidence, limitations & official sources
ISO/IEC 27001 information security management
Organizations establishing an information security management system.
Status: ISO/IEC 27001:2022 information-security management-system standard
Duties, evidence, limitations & official sources
ISA/IEC 62443 series - industrial security (4-2 component reference)
Industrial automation/control systems, with different parts addressing asset owners, service providers, system design and components. The linked IEC62443-4-2:2019 concerns technical security requirements for IACS components.
Status: Published guidance or standard; binding only where adopted
Duties, evidence, limitations & official sources
OWASP AI security guidance
Teams threat-modeling, testing and operating generative and agentic applications.
Status: Published guidance or standard; binding only where adopted
Duties, evidence, limitations & official sources
MITRE ATLAS
Security teams analyzing adversarial techniques against AI-enabled systems.
Status: Published guidance or standard; binding only where adopted
Duties, evidence, limitations & official sources
OECD AI Principles
Governments and organizations using the OECD policy framework and definitions.
Status: Published guidance or standard; binding only where adopted
Duties, evidence, limitations & official sources
Web Content Accessibility Guidelines
Web content and applications; legal incorporation and required level vary by jurisdiction and contract.
Status: Published guidance or standard; binding only where adopted
Duties, evidence, limitations & official sources
European Health Data Space Regulation
Health-data access/exchange, electronic health record systems and secondary uses within the regulation's scope; roles and phase-in dates differ.
Status: Entered into force March 26, 2025; principal operational obligations phased mainly to 2029 and 2031
Duties, evidence, limitations & official sources
EU Clinical Trials Regulation
Clinical trials of medicinal products for human use within the regulation. AI-assisted trial activities remain within sponsor/investigator responsibilities.
Status: Applicable; trial authorization/data requirements apply
Duties, evidence, limitations & official sources
EU Law Enforcement Data Protection Directive
Competent authorities processing personal data for specified criminal-law enforcement purposes. This is a different regime from ordinary GDPR processing.
Status: National transposition applies
Duties, evidence, limitations & official sources
Federal Records Act and NARA records management
Federal agency records, including electronic records created by AI-supported official business; ownership and retention depend on records status and schedules.
Status: In force
Duties, evidence, limitations & official sources
Solicitors Regulation Authority professional standards
SRA-regulated solicitors and firms. AI service providers are not a substitute for the regulated professional's judgment and supervision.
Status: Binding for regulated solicitors/firms within scope
Duties, evidence, limitations & official sources
OWASP Top 10 for Agentic Applications 2026
Technical reference for behavioral conformance. Evaluate the sequence of plans, tool requests, retries, delegations and outcomes against an approved operating envelope. A plausible final answer is not evidence that the route taken was authorized.
Status: Community risk taxonomy, published December 2025; not legislation
Duties, evidence, limitations & official sources
MCP authorization specification (2026-07-28)
Technical reference for action interlocks. Place enforceable checks between a proposed operation and its side effects. Tool availability, a model request and transport-level access must not be mistaken for business authorization.
Status: Pinned specification edition; transport authorization is not business-action approval
Duties, evidence, limitations & official sources
OpenTelemetry GenAI span conventions
Technical reference for decision evidence. Create an independently protected record connecting the mandate, relevant inputs, versions, policy decisions, actions and observed outcomes. Make uncertainty and missing evidence explicit.
Status: Technical conventions; many GenAI attributes/operations remain Development stability
Duties, evidence, limitations & official sources
NIST SP 800-218A: AI-model secure development profile
Technical reference for lifecycle accountability. Keep use cases, dependencies, obligations, control owners and material changes under accountable governance from procurement through retirement. Repeatable evidence supports decisions; it does not transfer accountability.
Status: Final, July 2024; complements SSDF, not a complete deployer compliance standard
Duties, evidence, limitations & official sources
NCCoE: software and AI-agent identity and authorization
Technical reference for authority lineage. Keep the initiating person, each delegated agent and the executing service attributable. A handoff should narrow or preserve authorized scope, not silently create new authority.
Status: NCCoE concept paper, February 2026; not a final practice guide or mandatory standard
Duties, evidence, limitations & official sources
CISA / international partners: SBOM for AI minimum elements
Technical reference for lifecycle accountability. Keep use cases, dependencies, obligations, control owners and material changes under accountable governance from procurement through retirement. Repeatable evidence supports decisions; it does not transfer accountability.
Status: Recommendations, May 12, 2026; not a universal legal mandate; official indexed publication reviewed, full page access restricted
Duties, evidence, limitations & official sources
CISA: 2026 minimum elements for an SBOM
Technical reference for lifecycle accountability. Keep use cases, dependencies, obligations, control owners and material changes under accountable governance from procurement through retirement. Repeatable evidence supports decisions; it does not transfer accountability.
Status: Guidance, July 29, 2026; complements, rather than substitutes for, component testing; official indexed publication reviewed, full page access restricted
Duties, evidence, limitations & official sources
California SB813 - independent AI verification organizations
Establishes a framework for independent verification organizations assessing AI systems/models. Enactment is supported by the official signing announcement.
Status: Signed September 9, 2026; detailed operative requirements require chaptered-text validation
Duties, evidence, limitations & official sources
California AB1405 - AI auditor registry and independence
Creates an AI auditor registry and an independence/transparency framework, as described in the official signing announcement.
Status: Signed September 9, 2026; registry implementation and operative details require validation
Duties, evidence, limitations & official sources
California September 2026 AI oversight executive direction
Directs state work on independent oversight and recommendations for stronger frontier-AI safeguards.
Status: September 18, 2026 executive direction; further proposals are not blanket private-sector mandates
Duties, evidence, limitations & official sources
California Delete Act / DROP data-broker deletion system
Registered data brokers within the Delete Act's scope; not every business maintaining an internal AI dataset is a data broker.
Status: DROP regulations effective January 1, 2026; broker processing duties started August 1, 2026
Duties, evidence, limitations & official sources
NIST AI RMF critical-infrastructure profile - concept stage
AI-enabled capabilities used by critical-infrastructure operators.
Status: Concept note released April 7, 2026; not a final replacement for AI RMF 1.0
Duties, evidence, limitations & official sources
MY APPROACH / SEVEN CONCURRENT RESPONSIBILITIES
Make every boundary accountable.
Seven overlapping assurance domains connect delegated authority, trusted information, observable behavior, authorized actions, defensible evidence, containment and accountable change.
These are descriptive organizing labels, not new industry standards. They operate together, not as seven sequential phases. The independent control plane remains a specific architectural boundary. Open each domain for its original slide terminology and source crosswalk.
Davis & Associates Advisory, LLC · Professional advisory
Governance that translates into accountable decisions.
Explore practical oversight for AI systems, cyber operations and regulatory obligations. Start a conversation about your organization’s priorities.