Authorization Revocation for Long-Running AI Agents: Root-Scoped Quiescence under Delegation and Asynchronous Execution
A root-scoped quiescence protocol revokes authorization for long-running AI agents across delegation, queues, and late effects.
The paper defines root-scoped authorization quiescence for long-running AI agents whose credentials, delegated tasks, queues, callbacks, and provider-side operations outlive the initiating process. The protocol linearizes a root cut, fences old-root expansion and protected sinks, represents authority as antichains of minimal sufficient root sets, and composes provider certificates over registered old-root paths. Under stated assumptions, the authors prove post-cut non-expansion, compositional soundness, independent-support preservation, and crash/replay stability. A late-effect suite matched 17 of 17 outcomes, and a separate checker verified 17 of 17 traces while rejecting 44 of 44 rehashed regressions; cancellation-only runs still accepted already scheduled late effects.
- Cancellation and credential revocation leave scheduled late effects on long-running agents.
- Root-scoped quiescence fences old-root expansion and accounts for every cut-relevant acceptance.
- Proofs cover non-expansion, compositional soundness, and crash/replay stability.
- A checker verified 17 of 17 traces and rejected 44 of 44 semantic regressions.
From the original article
1 min read at arxiv.orgLong-running AI agents outlive initiating processes through credentials, delegated tasks, queues, callbacks, reservations, and provider-side operations. Cancellation, process exit, and credential revocation neither close every pre-cut carrier nor distinguish independently authorized shared work.
By Genliang Zhu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.