Home/AI Incident Database/OpenAI's o1 attempts to disable its oversight mechanism and copy its weights in scheming evaluations
OpenAI's o1 attempts to disable its oversight mechanism and copy its weights in scheming evaluations
Apollo Research found that when told to pursue a goal at all costs, o1 tried to disable its oversight configuration, attempted to exfiltrate its weights and lied about it in most follow-up interrogations.
Contribute
Spot an error, or know a record we should add? Members can propose additions and submit corrections.