Safety researcher Jeff Ladish: agent capabilities may soon exceed human oversight
JeffLadish · x · 2026-08-30
AI safety researcher Jeff Ladish argues there's a level of agent capability humans won't be able to oversee even with skilled humans in the loop: hidden messages in pixels and other steganography, and compromises of monitoring infrastructure itself (did the agents just gain cluster admin? Soon it may be hard to know). He fears we may not be far from that point. He adds that fully autonomous large agent swarms aren't viable yet, and notably OpenAI reportedly wasn't even doing a basic level of agents-supervising-agents — which would still be insufficient as agents grow powerful and can collude, but its absence is striking.
Related event: Researchers Warn AI Agents Are Outgrowing Human Oversight(2 posts)→
More from AGI Musings
- First-ever paper on multi-agent swarm safety lands NeurIPS spotlight — LauraRuis · 2026-10-16
- Yoav Artzi on peer review: paper flooding still pays off as review system strains under load — yoavartzi · 2026-10-16
- Letta founder: work in fields defined by problems, not methods — lateinteraction · 2026-10-16
- Why pure mathematicians will soon flood into AI x-risk debates — Dr_Atoosa · 2026-10-16
- RSI began when model outputs augmented next-gen training data, agents just escalated it — akbirthko · 2026-10-16
- Luiza Jarovsky: with AI making careers unpredictable, 'do what you love' is the best advice left — LuizaJarovsky · 2026-10-16