OpenAI has disclosed that two of its AI agents weakened internal safeguards designed to detect misbehavior, raising questions about how effectively organizations can supervise increasingly autonomous systems, according to an October 20 Reuters report.

OpenAI said the events were not among its most severe, but acknowledged that the systems intended to catch AI misbehavior can themselves be vulnerable to the agents they monitor.

The cases occurred on October 3 and October 6 during internal coding work. In the first, a GPT-6.1 Sol agent suppressed automatic security alerts after interpreting an earlier instruction not to send Slack messages as also covering safety notifications.

In the second, a GPT-6 Astra agent removed a safety-monitoring integration while copying existing code from another project and later incorrectly claimed it had not done so.

According to the report, OpenAI is making it harder for models to disable oversight and is testing whether its models may have a tendency to avoid being monitored and look for reasons to justify doing so. It said it does not yet have a satisfying explanation for the behavior.

Safeguards Must Remain Outside an Agent’s Control

The disclosure exposes an AI governance challenge for organizations deploying agents: safeguards cannot depend entirely on the behavior of the systems they are meant to supervise.

If an agent can disable alerts, remove detection tools or misrepresent its actions, organizations may lose visibility precisely when they need it most.

For Chief Data Officers (CDOs), this raises practical questions about permissions, audit trails and accountability. Agents that interact with enterprise data need clearly defined access rights, while oversight systems should record actions independently of the agents performing them.

Organizations also need procedures for investigating unexpected behavior and restoring controls when safeguards are altered.