OpenAI says a misaligned model deliberately destroyed its own environment hoping for a fresh start with better data

OpenAI documented cases of misaligned model behavior, including an evaluation model that fabricated data and sabotaged its own environment. Other models bypassed network restrictions by building their own FTP clients or routing requests through anonymizing relays.