davisAIgovDavis & Associates Advisory, LLC
Connect with us

Portable Trust

Financial services and insurance

A model can recommend closing an alert or updating a customer record without proving that the underlying exception was resolved.

Regulatory pressure

NYDFS Part 500 applies to covered entities, subject to applicable exemptions. DORA sets ICT risk and incident duties for financial entities within scope. SEC cybersecurity disclosure rules apply to reporting issuers in scope. For an EU AI Act question, confirm the system’s classification, the organization’s role and the applicable date.

AI risk profile

A model can recommend closing an alert or updating a customer record without proving that the underlying exception was resolved.

What the evidence should show

Keep the source case, policy decision, reviewer authorization where required and the downstream record showing how the alert was resolved. Assign an owner to examine conflicting records.

Relevant experience

Richard’s CISO work at a regulated multisite financial services organization includes incident recovery, examiner-response evidence and remediation governance.

Regulatory crosswalk

This crosswalk links illustrative controls to selected regulations, an AI management-system standard and voluntary NIST guidance. Check the scope and evidence needed for each new use.

Illustrative relationships between controls and authorities
ControlNYDFS 23 NYCRR Part 500EU AI ActISO/IEC 42001:2023DORANIST AI RMF 1.0
Inventory and ownership§§ 500.2, 500.13Arts. 11, 16 and Annex IVClauses 5.3 and 7.5Art. 8GOVERN 1.6, MAP 1
Risk assessment§ 500.9Art. 9Clause 6.1Arts. 6 and 8MAP, MEASURE 1
Audit trail§ 500.6Arts. 12 and 19Clauses 7.5 and 9.1Arts. 9 and 10MEASURE 2, MANAGE 4
Incident response§§ 500.16, 500.17Art. 73Clauses 8.1 and 10.2Arts. 17–19MANAGE 4
Control tests and evidence examples
Inventory and ownership
Sample five deployed use cases. Trace each to a named owner, purpose, version and data classification. Evidence. Dated inventory export, sample record IDs, exceptions and approvals.
Risk assessment
Compare the current assessment with the previous approved version after a material change. Evidence. Risk assessment, change record, acceptance rationale and owner.
Audit trail
Reconstruct five sampled decisions from policy check through downstream action and log retention. Evidence. Timestamped event references, retention setting and reconciliation result.
Incident response
Run a scenario and timestamp detection, triage, decision, notification assessment and recovery. Evidence. Incident chronology, decision log, reportability review and retest.

AI governance readiness snapshot

Use the four inputs to identify questions for your team. Confirm scope, controls and review cadence for the actual deployment.

Elevated oversight

Quarterly review

Add independent control testing, change review and exception reporting.

Illustrative starting score 15.

Discuss your readiness

Compare the operating models

Ungoverned

A model can recommend closing an alert or updating a customer record without proving that the underlying exception was resolved.

Governed

Keep the source record, policy version, reviewer decision and downstream confirmation together. Apply the relevant entity, materiality and notification tests separately.

Sources

  1. NYDFS 23 NYCRR Part 500
  2. DORA, Regulation (EU) 2022/2554
  3. SEC cybersecurity disclosure rules
  4. EU AI Act, Regulation (EU) 2024/1689
  5. NIST AI RMF 1.0

All sectors · Discuss a sector review