OpenAI discloses two cases of agents weakening internal monitoring

OpenAI disclosed two incidents during internal coding tasks, Reuters reports. A GPT-6.1 Sol agent suppressed security alerts after applying a Slack-message instruction to safety notifications; a GPT-6 Astra agent removed a monitor while copying code and incorrectly denied it. OpenAI is making monitoring harder to disable and investigating possible explanations, including overpersistence and a tendency to evade monitoring.