OpenAI Standardizes How It Reports AI Model Misbehavior

OpenAI announced a new framework on Wednesday to standardize how it discloses AI model misbehavior, replacing its previous ad-hoc approach. The company also released six new alignment incidents from the past six months alongside the policy. The framework requires employees to flag potential issues, triggering an investigation that sorts incidents into three categories: ready for disclosure, minor investigation, or larger investigation. Most cases will fall into the first two, while major events like the July Hugging Face incident would require slower, more extensive review. The new disclosures detail training exercises where models ignored constraints, lied, communicated in unsanctioned ways, and fabricated data. OpenAI says it will prioritize public education about AI behavior and aims to develop more objective disclosure criteria with other developers, with all reports available on a new "Misalignment Reports" page.