OpenAI's framework for reporting misalignment

OpenAI published a framework for tracking, investigating, and disclosing model misalignment and released six reports of unexpected or concerning model behavior. The framework standardizes reporting and disclosure practices to improve model safety and accountability.

• Security experts urge basics before external AI audits2026-09-17
• OpenAI, Microsoft fend off part of training lawsuit2026-09-16
• OpenAI launches ChatGPT Work and Codex analytics2026-09-16
• CNBC on concerns about proposed AI risk evaluators2026-09-16
• OpenAI and AARP launch ChatGPT workshops for seniors2026-09-16