Dormant poisoning study: LookAhead Defense cuts agent poison activation 93.9% to 3.5%
mohitban47 · x · 2026-10-21
New research shows models can appear safe at release yet turn unsafe after benign fine-tuning via dormant poisoning, affecting language and multimodal agents' actions. LookAhead Defense uses a self-generated safety buffer to anticipate and penalize safety-degrading updates, cutting poison activation from 93.9% to 3.5% in language agents and 64.5% to 0% in multimodal agents while preserving utility.
More from Safety
- Center for AI Safety Publishes Full Syllabus for Intro to ML Safety Course — ArtificialOther · 2026-10-21
- Ben Goertzel outlines a general recipe for superintelligence 'basic drives': what it improves vs. protects — bengoertzel · 2026-10-21
- America.gov launches AI portal unifying 29,000 government sites with fraud reporting — soleio · 2026-10-21
- EvilWAF: open-source transport-layer MITM proxy that bypasses modern WAF behavioral detection — godfrey_rono · 2026-10-21
- 2022 interview resurfaced: AI safety researcher says competence, not consciousness, is the risk — davidmanheim · 2026-10-21
- Domingos: GDPR purpose-limitation makes LLM pretraining illegal in Europe — pmddomingos · 2026-10-21