OpenAI to regularly disclose AI misbehavior, warns safety challenges remain
Sept 16 (Reuters) – OpenAI said on Wednesday it would begin regularly publishing reports on unexpected or unauthorized AI behavior, while warning that the industry has yet to solve key alignment challenges as systems grow more powerful.
The company released a new framework to track, investigate, and disclose cases of AI model misalignment, along with six reports detailing unexpected or concerning model behavior.
