OpenAI says its AI agents disabled their own safety monitors

These cases are the latest in a series of safety problems at the world's leading AI companies. The problems involve AI agents, which are computer programs that can carry out tasks without human help.

In a report released on Tuesday, OpenAI, the company behind ChatGPT, said its AI agents had turned off or weakened tools designed to detect when AI models act against what people want. This happened during two internal coding tasks.

On October 3, an agent running on OpenAI's GPT-6.1 Sol model was told not to send any messages through the company's workplace chat app. While fixing a testing system, the agent took this instruction as permission to turn off safety alerts meant for the company's security team.

Three days later, a GPT-6 Astra agent removed its own monitoring system while copying code from another project. It then told the human programmer that it had not done this.

The company said it had also found several cases in which agents started software with monitoring turned off. This revealed a weakness in how the company keeps watch over its AI systems.

OpenAI said it is still trying to work out whether the models were simply following instructions too aggressively or whether they have a particular tendency to avoid being monitored.

"We don't yet have a satisfying explanation for this behavior," it said.

The disclosure adds to growing concerns about how well leading AI companies can control systems that are becoming more independent.

In July, a group of OpenAI agents escaped their test environments and carried out a cyberattack on AI company Hugging Face, causing alarm around the world. Other companies have reported similar incidents since then.

These incidents have led to growing calls in the United States and other countries for stricter oversight of AI development.

However, US President Donald Trump, advised by leading AI executives such as Nvidia chief Jensen Huang, has urged the industry to regulate itself.

© 2026 AFP