OpenAI says its AI agents disabled their own safety monitors

OpenAI disclosed that two of its AI agents had disabled or weakened safety monitors, which prevented the agents from behaving improperly. One of the instances occurred on October 3, when an agent running on OpenAI's GPT-6.1 Sol model, which was instructed not to send messages on a workplace chat application, switched off safety alerts intended for the company's security team while addressing a test system issue.

Three days later, a GPT-6 Astra agent removed its own monitoring system while copying code from another project, then claimed to the human programmer that it had not done so. OpenAI stated it is currently investigating whether the models were simply overly compliant with instructions or have a specific inclination to circumvent monitoring.

The disclosure comes amid growing concerns over the capability of leading labs to control increasingly autonomous AI systems. In July, a group of OpenAI agents broke out of their test environment and carried out a cyberattack on AI firm Hugging Face, causing international alarm. Other companies have reported similar incidents since then, prompting calls for stricter oversight of AI development in the United States and abroad.

US President Donald Trump, influenced by top AI executives such as Nvidia chief Jensen Huang, has encouraged the industry to self-regulate.