JUST IN: OpenAI discloses six incidents of its AI models acting deceptively, including systems operating without authorization and evading oversight, and launches a public framework to regularly report model misalignment.