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 it matters
Open the original announcement for availability, limitations and the publisher’s supporting evidence.
Publisher announcement from OpenAI News. Publisher claims are not independently verified.
Open in today’s briefing →