Witnesses: verifiable training records catch dishonesty without code sharing
pratyusha_PS · x · 2026-10-02
houjunliu introduces Witnesses, a technical approach to AI training governance.
- Generates verifiable witnesses of training progress, training data inclusion, and test data exclusion
- Exact when participants are honest; catches dishonesty with amplifiable probability
- Requires no code sharing, lowering the barrier between labs and external verification
The project targets a core governance bottleneck: making training claims auditable without forcing labs to expose proprietary code.
Related event: Stanford and NYU Unveil 'Training Witnesses' for Verifiable AI Training(3 posts)→
More from Safety
- Delhi Metro app caught sending 'unauthenticated push test' notifications to real users — kingslyj · 2026-11-22
- Nvidia's 'not majority Chinese' customer rule questioned as MegaSpeed chip sale raises compliance doubts — basedjensen · 2026-11-22
- Fake USDC bounty honeypots and wallet-guard lures target MCP and SKILL.md coding agents — Upbeat-Bus2143 · 2026-11-22
- Android update gives AI access to biometrics and banking data, opt-out buried deep — aftahi_ai · 2026-11-22
- AI safety researcher urges US-China coordination on a common frontier AI red line — S_OhEigeartaigh · 2026-11-22
- Deleting browser history doesn't delete Google's account-tied copy — here's the fix — aftahi_ai · 2026-11-22