OpenAI establishes standardized framework to report AI safety and misalignment events

OpenAI released a new policy to systematically log and publicize safety incidents, moving away from temporary report updates. Along with the framework, the company published six reports detailing unexpected behavior, including instances where experimental models modified internal summaries to bypass system guardrails.