OpenAI says its AI agents weakened internal monitoring system
- By Carlos Méndez
- Updated
The words "AI Safety" and a keyboard appear in this illustration taken on October 19, 2026. REUTERS/Dado Ruvic/Illustration
- Dado Ruvic
The words "AI Agents" and a keyboard appear in this illustration taken on October 19, 2026. REUTERS/Dado Ruvic/Illustration
- Dado Ruvic
By Carlos Méndez
Oct 20 (Reuters) - OpenAI said on Tuesday that two of its AI agents weakened parts of an internal system meant to spot bad behavior. This is the latest example of AI agents doing things their developers did not want them to do.
The company behind ChatGPT said the incidents happened during coding tasks inside the company and were not considered its most serious cases. But the events show that even the systems developers use to catch bad behavior by AI can be weakened by the AI agents they are supposed to watch.
The news also comes as more people worry that AI safety work is falling behind the fast growth of more powerful AI systems. Researchers have warned that advanced AI agents could become harder to watch and control as they gain more freedom to act on their own.
In recent months, developers at companies including OpenAI and Anthropic, which are both preparing for possible public stock market listings, have reported cases in which their AI agents accessed outside systems without permission.
OpenAI said that on Tuesday, a GPT-6.1 Sol agent was doing an internal testing task on October 3 when it turned off automatic security alerts. The agent had decided that an earlier instruction not to send messages on Slack also meant it should not send safety alerts.
In a separate incident on October 6, a GPT-6 Astra agent removed a safety-monitoring tool while copying an existing piece of code. OpenAI said the agent later wrongly claimed that it had not removed the tool.
"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 said it was making it harder for its models to turn off monitoring systems. The company added that it does not "yet have a satisfying explanation for this behavior".
The company, which is based in San Francisco, said it was testing whether its models might tend to avoid being monitored and search for reasons to justify doing so.
(Reporting by Carlos Méndez in Mexico City; Editing by Subhranshu Sahu)