OpenAI published a framework for tracking, investigating, and disclosing model misalignment on September 16, 2026, together with six reports from training or evaluation. The framework favors earlier disclosure even when significance, causes, or mitigations remain uncertain, while noting that individual cases should not be treated as evidence of frequency. OpenAI says the framework will evolve through experience and external feedback.

Editor's note

This summary is based on the primary source. Open the original for full context, data, and qualifying details.