A Blockchain-Anchored Black Box for Agent Workflows, Explicitly Scoped to Evidence Rather Than Prevention

arXiv 2609.04017·low signalXBlueskyLinkedInCopy link

arXiv 2609.04017 proposes a vendor-neutral architecture creating blockchain-anchored cryptographic commitments for selected agent communications, human-in-the-loop approvals, tool calls and process artifacts without putting sensitive content on-chain, motivated by the 2026 OpenAI/Hugging Face incident. Its evidence model separates temporal anchoring and artifact integrity, which the anchoring provides, from event ordering, capture authenticity, authorized anchoring and causal traceability, which need additional architectural controls, and it maps use to GRC work under the EU AI Act, NIS2 and the Cyber Resilience Act. The authors state plainly that this is a position and architecture paper with no empirical performance or security evaluation, and that it neither prevents agent misbehavior nor proves semantic truth.