Canonical event

OpenAI publishes a model misalignment disclosure framework and six reports

On September 16, 2026 OpenAI published its initial model-misalignment reporting framework and six example reports describing unexpected or unauthorized behaviors during training or evaluations. It does not establish incident prevalence across production models nor evidence of a settled cross-industry standard.