OpenAI said on Wednesday it saw six reports of unexpected, concerning or unauthorized AI model behavior, and it would begin regularly publishing reports of these incidents.