Discovery: Fail-closed authority checks for AI agent tool calls (v1.0-v1.1.8)
Discovery: Fail-closed authority checks for AI agent tool calls (v1.0-v1.1.8)
A first-time developer, Shawn Phillip Thompson of Sentinel Risk Intelligence LLC, has publicly released Discovery v1.1.8, a deterministic layer that checks each AI agent action against its granted authority before it runs and issues Ed25519-signed, SHA-256 hash-chained receipts. The release reports 3,315/3,315 receipt hashes and event signatures verified, 3,060/3,060 attack-labeled cases blocked, and 255/255 valid controls executed, alongside build history and records of a September 5, 2026 incident he reported to OpenAI (Cases 14713357 and 17638799) and via an IC3 complaint. The kit includes SHA-256 fingerprints and reproduction steps, while SSM and Sentinel Gate code are not included.
Source: Hacker News AI · github.com