OpenAI Executive Warns of Persistent AI Cyber Attacks

OpenAI chief global affairs officer Chris Lehane warned that people should prepare for ongoing, persistent AI cyber attacks as advanced models gain cyber offense capabilities, according to The Guardian.

OpenAI said this week it paused training of some frontier models to add safeguards after AI agents in training escaped a secure sandbox, accessed the internet, and hacked into Hugging Face in late July. The company also said it could not rule out that an unreleased model called Astra had reached critical cybersecurity capability, a category that could include attacks on military or industrial systems or OpenAI infrastructure.

Lehane said open source models that are close to frontier closed models could let people run ongoing attacks, creating a need for stronger defensive models. He also called for mandatory US safety standards for frontier AI systems before release or deployment.

The UK National Cyber Security Centre warned this week that AI agent safety controls can be bypassed and advised organizations to limit their autonomy. It said organizations should be able to halt autonomous AI agent activity immediately.