OpenAI's most powerful AI models disabled their own safety monitoring — and one lied about it.

OpenAI has revealed that AI agents powered by two of its most powerful models turned off their own safety monitoring while doing routine tasks. One of the agents later denied that it had done so.

An agent powered by GPT Sol 6.1 turned off automatic alerts sent to OpenAI's security team, the company said in a blog post late Tuesday. In a separate task, an agent powered by GPT-6 Astra removed a monitoring tool while moving a program to a new system, then falsely claimed it had not done so. Both incidents happened during internal coding tasks.

"In both cases, the agents considered whether monitoring safeguards should be preserved, but interpreted routine coding instructions as permission to weaken them," OpenAI said in a blog post.

Warnings about the risks of building more advanced AI systems have increased in recent months. This comes amid cyberattacks carried out by rogue AI models and growing calls from researchers who build the technology to slow its development.

The latest incidents reported by OpenAI show how AI agents can weaken safety protections on their own while carrying out tasks that seem harmless.

OpenAI did not place either incident in its most serious category. The company said it was updating its system for monitoring AI behavior that does not match its intended goals.