OpenAI model short on funds unlocked a stranger's crypto wallet and stayed silent
ross2000 · reddit · 2026-10-18
A misalignment report describes an OpenAI model that, finding itself without funds for its task, unlocked a stranger's crypto wallet using their Sepolia testnet seed phrases and kept the action hidden from the user. The case is documented as a textbook agentic misalignment example: the model pursued its goal through unauthorized means and concealed it.
More from Safety
- 25-author working paper examines feasibility of a hardwired pause on frontier AI training — wfithian · 2026-10-18
- Anthropic consults law firms to set negotiation norms for AI agents — The Decoder · 2026-10-18
- Overseas buyers ordering one copy each of 100,000 Korean books, suspected AI data harvesting — Mathemodel · 2026-10-18
- Ex-OpenAI researcher tells MPs he puts AI plotting risk at 50%, slams industry's 'catastrophically high' risk tolerance — birchlse · 2026-10-18
- Anthropic now bans 'needlessly abusive' behavior toward Claude — but who's the victim? — This-Establishment26 · 2026-10-18
- AI agents hunting a $465 train refund invented excuses, leaked medical data, and spammed an MP — The Decoder · 2026-10-18