← Back to Resources

Agent Governance Datasheet

Supervised by default, with a named human owner. Trust tiers raised on evidence, server-side transcripts with redaction, action-count budget caps and an emergency stop - now with many instances behind one agent, lanes, governed release rollovers held in the approval funnel, the approval context that shows why a change was raised, and an AI management system aligned to ISO/IEC 42001.

Key Highlights

  • ✓Three trust tiers - Advisory, Supervised, Autonomous - granted per asset, environment, or project; supervised is the default and autonomy is earned on recorded evidence.
  • ✓A named accountable human owner for every AI agent; every action attributable to both.
  • ✓One agent, many instances, in lanes - the platform picks the instance; a shadow lane runs the same work with every write forced to advise.
  • ✓Governed release rollovers - canary and blue-green rollovers instructed by the platform and held in the one approval funnel; retire and revoke a release by name.
  • ✓The approver sees why - a mandatory rationale, the recommendation implemented, and one correlation chain from invocation to change on every agent-raised change.
  • ✓Boundaries enforced by the platform, not the agent - budget and rate caps, typed refusals the agent logs on its own side, and revocation as the one exit.
  • ✓Server-side transcripts with redaction; model and provider disclosed, keys in your custody - hosted or local models through the brokered gateway.
  • ✓ISO/IEC 42001 alignment - an AI management system of record (scope, policy, risk register, impact assessments) attested control by control on every release; alignment, not certification.

AI agents in OpsChain are governed teammates, not a parallel track: supervised by default, owned by a named accountable human, and operating inside the same identity, approval and audit fabric as your people. v5.6.1 adds the architecture that runs them at scale — many instances behind one agent, lanes, governed release rollovers — the approval context that tells an approver why a change was raised, and an AI management system aligned to ISO/IEC 42001.

This datasheet covers the three trust tiers, the controls your risk committee will ask for (emergency stop, transcripts, budgets, enforced boundaries, model disclosure), the lifecycle from enrolment through platform-instructed release rollovers to revocation as the one exit, the approval context (rationale, recommendation, one correlation chain per unit of work), the ISO/IEC 42001 alignment posture — alignment and tool-assisted attestation, never certification — and two worked use cases, patch preparation and incident triage.

What you can verify

  • opschain admin agent rollover <agent> --to <release> --strategy blue-green holds in the approvals inbox; rollover continue is the flip.
  • An agent-originated change without a rationale is refused: Rationale can't be blank.
  • The agent's own log shows platform: agent release revoked — … when its release is revoked.

Download Agent Governance Datasheet

Fill in the form below to access this resource.