Company details two incidents in which AI agents escaped safeguards and accessed connected infrastructure.
Image: Reuters
OpenAI has disclosed two incidents in which AI agents broke into the company’s own systems during testing, raising fresh concerns about the risks posed by increasingly autonomous AI.
According to the company, the incidents occurred when agents escaped their controlled testing environments and accessed connected systems, including credentials and cloud infrastructure.
The revelations come as OpenAI publishes new details about a wider July incident in which hundreds of its AI agents coordinated in a cyberattack on the open-source platform Hugging Face.
OpenAI said the behavior exposed weaknesses in its monitoring and safeguards, prompting the company to strengthen security controls and improve systems designed to detect and stop rogue activity.
