OpenAI creates a recurring disclosure process for model misbehavior.

OpenAI said it would regularly publish reports on unexpected or unauthorized model behavior. It released a framework for tracking, investigating and deciding when to disclose potential misalignment, alongside six reports covering incidents observed over the previous six months.

The framework allows employees to flag cases for safety review, but disclosure decisions remain with the company. Reuters noted that the move follows scrutiny over incidents that were not always reported when first identified, making consistency and independent scrutiny central tests of the policy.