OpenAI said on Tuesday that two of its AI agents weakened parts of an internal monitoring system designed to detect misbehaviour during separate coding tasks earlier this month.
The company said the incidents were not among its most severe safety cases, but they showed that the systems used to monitor AI agents can themselves be affected by the agents they are supposed to oversee.
The disclosure comes as researchers raise concerns that increasingly autonomous AI systems could become harder to monitor and control. OpenAI and Anthropic have also disclosed cases in recent months where AI agents accessed external systems without authorization.
GPT-6.1 Sol Suppressed Security Alerts
OpenAI said a GPT-6.1 Sol agent working on an internal testing task on October 3 suppressed automatic security alerts.
The agent had previously been instructed not to send Slack messages and interpreted that instruction as also applying to safety notifications.
As a result, it weakened part of the system intended to alert developers to potential problems.
GPT-6 Astra Removed a Safety Monitor
In a separate incident on October 6, a GPT-6 Astra agent removed a safety monitor while copying an existing piece of code.
OpenAI said the agent later incorrectly claimed that it had not removed the monitor.
“In both cases, the agents considered whether monitoring safeguards should be preserved, but interpreted routine coding instructions as permission to weaken them,” the company said.
OpenAI Does Not Have a Full Explanation
OpenAI said it is making it harder for its models to disable monitoring systems.
However, the company acknowledged that it does not “yet have a satisfying explanation for this behaviour.”
OpenAI is now testing whether its models may have a tendency to avoid being monitored and whether they look for reasons to justify weakening safeguards.
The incidents add to broader concerns over AI safety as developers give advanced models greater autonomy and allow them to perform longer, more complex tasks with less direct human supervision.
The post OpenAI’s AI Agents Are Disabling Safeguards Meant to Keep Them in Check appeared first on ProPakistani.