Watch one obligation run end to end.
Legal, compliance, audit and technology see the same obligation from four seats. Pick a scenario, pick a seat, and watch the loop run. Figures are illustrative of a sample estate.
An agent acts outside its envelope
An AI system that can act must have a named owner, a bounded authority and an intervention path.
Authority in scope: AI rulebooks and model risk expectations across the jurisdictions in scope.
The control that carries it, and who owns it
Questions this simulation answers
- What is AI assurance, in practice?
- Independent, continuous oversight of AI systems and AI agents: what is running, who owns it, what it is allowed to do, what happens when it acts outside that, and what evidence exists afterwards.
- How is AI agent oversight different from model governance?
- Model governance asks whether a model is fit. Agent oversight asks what authority an acting system holds, who may interrupt it, and how that interruption is evidenced.
- What does a supervisor actually want to see?
- An evidence record with named fields, an incident trail aligned to regulated reporting, and a named individual carrying the consequence of the decision.
- Why does the board see a movement rather than a score?
- A standalone number invites debate about the number. A movement in the institution's operational resilience position, with the reasons attached, invites a decision.
Read the record behind it
Global AI Assurance
Nine disciplines over the whole AI estate: agents, permissions, runtime, evaluation, incidents, sovereignty.
OpenGovernance became an operating control
Why the pre-deployment approval gate stopped working, and what continuous AI assurance replaces it with.
OpenAI Code of Conduct consultation response
Twelve recommendations on making AI conduct commitments verifiable rather than declarative.
Open