Problem and Fit
Should this responsibility belong to an AI system?
Business outcome, engineering behavior, AI suitability, risk class.Signature Framework
The review before an AI system gains power.
A principal-level gate for adding autonomy, exposing users, connecting tools, or increasing the system's blast radius.
System Boundary
What does the system own, and what remains outside it?
Inputs, outputs, users, policies, context sources, tools, humans, downstream systems.Decision Surface
What may the AI inform, draft, recommend, route, decide, or act on?
Autonomy level, approval rules, abstention conditions, escalation paths.Blast Radius
How far can one failure propagate?
Affected users, records, transactions, data exposure, cost, trust, reversibility.Evidence and Release Gate
What evidence is enough before exposure increases?
Evaluation thresholds, security review, rollout plan, rollback path, operational signals.When to run it
Run the framework whenever power increases.
- adding tools or write access
- increasing autonomy
- expanding user exposure
- changing model, prompt, context, or routing
- moving from pilot to production
- increasing blast radius or consequence