Our framework for reporting model misalignment
OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.
Why this matters
This briefing preserves the publisher-provided context in a clean, searchable format. Use the original report for the complete announcement, technical details, evidence, and any subsequent updates.
READ THE ORIGINAL REPORT ↗