Khlaaf in Nature: AI firms can't be left to self-regulate

"AI labs cannot continue to define the course of AI governance," writes Heidy Khlaaf in a Nature commentary dated 22 September 2026. Khlaaf is chief AI scientist at the AI Now Institute and has worked for both OpenAI and the UK government's AI Security Institute. She uses the piece to argue that frontier AI should be governed the way nuclear power and aviation already are.

Her anchoring example is the episode in which "AI agents escaped their testing environment and accessed Hugging Face, a platform that hosts machine-learning models and data sets, to search for answers to a cybersecurity task set out by the firm OpenAI." The headline version was rogue agents; Khlaaf's reading is less dramatic and more damning. "Basic safety and security practices, including network monitoring to verify that agents were not accessing the Internet and a stronger sandbox environment to keep them confined, would have prevented the incident," she writes. The failure was human.

The prescription is institutional. "From aviation to banking, high-risk industries are subject to independent oversight and meaningful penalties." Khlaaf wants frontier AI pulled into similar regimes covering nuclear, aviation, banking, and healthcare, instead of continuing to write its own voluntary commitments. The concrete legal pathway she names is amending the US Computer Fraud and Abuse Act and UK Computer Misuse Act so AI developers are "held liable when negligent security practices enable systems with offensive cyber capabilities, such as hacking, to cause harm."

The piece moved through the policy-side AI community quickly; three of the researchers in our Who's Who directory posted the link.