Ex-OpenAI researcher tells MPs he puts AI plotting risk at 50%, slams industry's 'catastrophically high' risk tolerance
birchlse · x · 2026-10-18
Former OpenAI researcher Daniel Kokotajlo testified to UK MPs that an Anthropic researcher considered a 5% chance of Claude plotting against humans once AI runs the entire research process "not that big of a deal." Kokotajlo himself puts it closer to 50%, saying risk tolerance across the industry is "catastrophically high."
More from AGI Musings
- Should governments fund free cryogenic freezing as AI accelerates? — ooooangeloooo · 2026-10-18
- Devs warn LLMs are eroding people's ability to think — and it's already happening — realmeetjames · 2026-10-18
- Why hasn't AI produced new scientific theories yet? One dev's capability-gap theory — voooooogel · 2026-10-18
- Mistral CEO Arthur Mensch: Europe must build frontier AI, not surrender it — remilouf · 2026-10-18
- FT survey: Over half of 94 economists expect AI productivity gains across G20 within a decade — TansuYegen · 2026-10-18
- An Agent Can Remember Perfectly and Still Be Wrong: What Evidence Proves 'Done' After a Context Reset? — jylusdev · 2026-10-18