Observed arrival · 2026-11-07
AI VERIFY: Proof of AI Work
A demonstrator for checking AI-agent outputs against explicit requirements and captured evidence, then returning PASS, FAIL, or UNVERIFIED.
- For
- Teams reviewing outputs from AI agents
- Worth noticing
- The page reports 28 matching labels in a 30-case comparison; two SHA-256 cases are unsupported and return UNVERIFIED.
Field notes
Operators register requirements and trusted reference data outside the agent's control. Captured outputs are checked by deterministic predicates—JSON equality, CSV sums, SHA-256 matches, and file existence—without executing submitted programs. The browser simulation uses recorded fixtures and recomputes only simple checks locally; it does not verify cryptographic signatures or host authentication. The page reports 28 matching labels in a 30-case comparison, while two SHA-256 cases remain unsupported.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue