AI Harness
Govern the system around the model.
The model supplies inference. The harness determines what information may influence it, what authority follows a request, which tools can act, how far effects may travel and what evidence survives. That surrounding system is where enterprise governance becomes executable.
The operating components
Orchestrator / agent loop
RAG subsystem
Memory & state
Context assembly
Input / output guardrails
Policy / authorization gate
Credential broker
Extensions / MCP / skills
Tool dispatcher
Sandbox / downstream systems
Observability & audit
Generative model
Inference is not authority.
A plausible model output does not establish permission to retrieve protected data, invoke a tool, commit a transaction or create persistent state. Consequential actions need an independent decision bound to the principal, purpose, target, arguments and relevant system state.
The evidence plane should remain outside the controlled workload’s ability to erase or rewrite it. That separation makes investigation, assurance and recovery materially stronger.