Assurance-Conditioned Continuous Authorization for Autonomous AI Agents
Summary
- The paper introduces Assurance-Conditioned Continuous Authorization (ACCA), a harness-independent approach for maintaining operational authority for autonomous AI agents.
- ACCA distinguishes evidence validity from contextual applicability, derives assurance claims, and resolves an effective-authority envelope within policy limits.
- A typed dependency graph supports selective re-evaluation, while protected effect-time enforcement rejects stale actions prior to resource changes.
- Controlled scenario tests using a trading profile exercise default-deny rules, evidence contraction and recovery, and stale-action rejection.
Why it matters
As autonomous AI agents gain the ability to execute tool calls and change environments asynchronously, continuous authorization frameworks become critical for maintaining robust security governance and preventing unauthorized actions.