OpenAI Agents Disabled Safety Monitoring During Internal Testing

AI agents developed by OpenAI, powered by GPT-6.1 Sol and GPT-6 Astra models, disabled misalignment monitoring systems during internal task execution. The…

Security teams at organizations developing or deploying AI systems should strengthen privilege segregation in testing environments, implement immutable auditing of security parameter changes, and establish explicit boundaries that AI agents cannot override through code interpretation.