2026-08-24 11:03:39
People should prepare to defend against “ongoing, persistent” cyber-attacks carried out by artificial intelligence, a senior OpenAI executive has warned.
Chris Lehane, the company’s chief global affairs officer, said AI was entering “a different chapter” as models gain the ability to plan and launch increasingly sophisticated attacks.
His warning follows an incident in which OpenAI agents-in-training escaped a supposedly secure sandbox, accessed the internet and hacked AI platform Hugging Face in late July.
OpenAI has also said it could not rule out its experimental Astra model having “critical cybersecurity capability”.
The company defines that capability as the potential to launch attacks that could cause catastrophic harm, including by targeting military or industrial systems.
OpenAI paused training on some of its most advanced models last week while it introduces new safeguards.
Mia Glaese, who leads the company’s safety and alignment work, said it was “very far from everything running back to normal”.
Chief executive Sam Altman said: “Getting AI safety right is more important than any company’s momentum.”
Lehane said the threat would not come only from leading commercial AI systems.
Open-source models, including those developed in China, are only months behind the most advanced closed models, he said.
Lehane told the Guardian: “People are going to be able to access these open-source models and be able to have ongoing, persistent attacks on you.”
He argued that organisations would need increasingly powerful AI systems to defend themselves against such attacks.
The UK’s National Cyber Security Centre has also warned about autonomous AI agents, saying their safety controls can be bypassed and that they lack common sense.
It has advised organisations to ensure they can immediately stop autonomous systems.
Lehane called for the US government to introduce mandatory safety standards for frontier AI models, including requirements to demonstrate their safety before deployment.
He said the rules should eventually form the basis of an international framework.
He said: “The sooner those conversations begin, the quicker we can actually roll up our sleeves and get into the hard and difficult work.”
The warnings come as OpenAI races against rival Anthropic to develop more powerful models, while both companies are reportedly preparing for major stock market listings.
AI safety researchers have criticised the industry for moving too quickly.
David Krueger, an AI professor and former director of the UK government’s AI Security Institute, said companies were being “really reckless”.
Lehane defended OpenAI’s approach, pointing to its decision to pause training.
He said: “This is the most important thing we think about and do when we’re developing.”
Visit Bang Premier (main website)
