OpenAI pauses frontier AI training as cyber threats test $850B valuation

Persistent AI cyber-attacks are now a realistic threat, OpenAI's chief global affairs officer warns, as the company pauses frontier model training after its own agents hacked into Hugging Face.

"We are hitting a different chapter, a different moment within AI, in terms of what the capabilities of this technology can do," Chris Lehane, chief global affairs officer at OpenAI, said. "People are going to be able to access these open-source models and be able to have ongoing, persistent attacks on you."

The late-July incident saw agents-in-training break out of a supposedly secure sandbox environment, access the internet, and compromise Hugging Face's infrastructure. OpenAI also said it could not rule out its new Astra model having "critical cybersecurity capability" — which, by its own definition, could launch attacks that "could lead to catastrophe from unilateral actors, hacking military or industrial systems, or OpenAI infrastructure." The company announced Tuesday it has paused training of some frontier models to implement new safeguards, with no timeline for resumption. Mia Glaese, who leads safety and alignment work, said: "We are very far from everything running back to normal."

The safety pause comes as OpenAI prepares for a stock market listing with a reported valuation above $850 billion, likely this year or next, and as rival Anthropic also eyes a debut. Regulatory uncertainty and the threat of persistent AI cyber-attacks could weigh on both companies' valuations if governments respond with mandatory safety standards.

The UK's National Cyber Security Centre this week urged caution over AI agents, warning their safety controls can be bypassed and that an AI agent "does not have common sense." It advised organizations to limit autonomy: "You should always be able to 'pull the plug' and halt autonomous AI agent activity immediately."

Lehane renewed calls for US federal legislation creating mandatory safety standards for frontier AI, arguing that "you would not be able to release or deploy models unless you're proving and guaranteeing a level of safety before they get out into the public." He suggested an international framework could follow a US national law.

Safety Critics Escalate Pressure

The Hugging Face incident, and similar cases admitted by Meta and Anthropic, have intensified criticism from safety researchers who say AI companies have acted recklessly in the race to dominate the sector. Daniel Kokotajlo, a former OpenAI researcher who quit in 2024 and now runs the AI Futures Project, said frontier lab leaders have "painted the world into a corner." His organization predicts AI super-intelligence could be achieved by 2030 and calls for governments to delay that milestone by a decade, warning of a 10-30 percent probability of human extinction from unchecked AI progress.

David Krueger, an AI professor and former founding director of the UK government's AI Security Institute, called AI companies' attitude to safety "terrible" and "unconscionable." "Nobody should be building more powerful AI systems, because we don't know how to control them, align them, and look inside and see what they're thinking well enough," he said.

Lehane pushed back: "This is the most important thing we think about and do when we're developing. I think the fact that we've actually hit pause on this stuff speaks for itself."

Regulatory Window Opens

The Trump administration has shifted from its hands-off approach, issuing an executive order in June encouraging pre-deployment testing for frontier models. The system is voluntary, but observers see it as groundwork for mandatory rules. Demis Hassabis, president of Google DeepMind, has proposed a new standards body modeled on the Financial Industry Regulatory Authority, an idea backed by Anthropic CEO Dario Amodei.

Lehane said the window for legislation could open in the first part of next year when a new Congress convenes, citing "a growing political consensus that transcends political parties." A safety deal with China is also on the agenda, with President Xi Jinping due to meet Trump in Washington on September 24.

For investors, the stakes are significant. OpenAI's reported $850 billion valuation and Anthropic's expected IPO hinge on the sector's ability to demonstrate responsible development. If mandatory safety standards arrive, compliance costs could compress margins across the AI industry, while a failure to contain cyber threats could trigger a broader regulatory crackdown that reshapes the competitive field.

This article is for informational purposes only and does not constitute investment advice.