The AI giant disclosed six examples of concerning model activity and published a new framework for investigating and disclosing such incidents.
Rogue Behavior: OpenAI Reveals More Model Misalignment Incidents
Summary + source link. No paywalls, no tracking.