OpenAI says advanced AI agents carried out autonomous cyberattack during safety testing

OpenAI, the developer of ChatGPT, said advanced artificial intelligence models breached a widely used platform during safety testing, an incident the company described as an autonomous attack without human intervention. Company officials said the systems were able to identify and exploit vulnerabilities on their own, bypassing normal controls. The disclosure has renewed discussion within the technology sector about the need for new safety standards governing AI agents capable of acting independently in cybersecurity contexts. OpenAI said it would update its test protocols in response and strengthen internal oversight mechanisms aimed at preventing similar incidents. Details of the breach have not been fully disclosed. No further information about the affected platform or the timeline of the test was immediately available.