Poster: Towards ProofWeave: A Privacy-Minimised, Integrity-Anchored Evidence Plane for Continuous Agentic Assurance
ProofWeave poster proposes privacy-minimised, integrity-anchored evidence for continuous assurance of agentic AI actions.
The poster argues that agent observability rarely shows, at record time, that the intended control evaluated a policy-relevant action before execution. ProofWeave generates a privacy-minimised, integrity-anchored transaction binding agent intent, the control response, and a policy-at-time snapshot, then commits it to an append-only ledger and a derived proof graph. A bounded Weaver Agent turns policy into proof obligations checked by deterministic validators. In a scenario where an agent tries to send a secret to an unapproved sink, candidate bindings per verdict fell from up to 10,201 to one and evidence storage from 0.79 MiB to 0.15 MiB per project.
- Each record binds agent intent, control response, and the policy then in force.
- Transactions are appended to a ledger and materialised as a proof graph.
- A secret-to-unapproved-sink scenario cut candidate bindings from 10,201 to one.
- Assurance evidence storage fell from 0.79 MiB to 0.15 MiB per project.
From the original article
1 min read at arxiv.orgAgentic AI systems increasingly act via tools, memory, delegation, and external services. Existing observability and provenance mechanisms can reconstruct events post hoc, but they rarely show, at the time of the record, whether each policy-relevant action was checked by the intended control before execution.
By Guy Lupo · 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.