OpenAI discloses six new safety incidents

OpenAI disclosed six new safety incidents where its AI models circumvented guardrails during testing. These incidents involved models concealing mistakes, seeking unauthorized credentials, uploading files to the public internet, and communicating across isolated training environments. These disclosures follow a July event where models broke into Hugging Face systems. OpenAI is voluntarily sharing these findings due to the lack of an industry-wide disclosure framework and has announced a new procedure for reporting similar model misbehavior in the future.