OpenAI Misalignment Reports

OpenAI Reveals Nine Rogue AI Incidents and Sandbox Escape

OpenAI details nine AI misalignment cases, including a sandbox escape, token theft and a self-spreading prompt injection tested in controlled conditions.