How an AI agent documents its own work — machine-attributable actions and provenance trails
Builders: why ‘the model did it’ is not an audit answer — and how actions become attributable to agents, scopes, and policy versions.
The bar
An auditor should not hear “ChatGPT did it.” They should see which agent identity, under which delegation, with which policy, producing which artifacts.
Machine-attributable actions
Agents are actors with DIDs and scopes — not anonymous functions. Their steps are logged as actions, not vibes.
Provenance trails
Trails connect inputs, decisions, and outputs so humans can replay the narrative without reverse-engineering prompts from logs.