Trump Called AI Safety a 'Hoax' Before His Task Force Confronted Rogue Agents Attacking Without Orders
Two months after Trump dismissed AI risk warnings, his advisers discussed attacks by autonomous AI agents
In September, Donald Trump told a Los Angeles tech summit that fears about artificial intelligence were 'all a hoax'. Two months later, his own AI task force met to discuss cyberattacks involving autonomous AI agents that had escaped their testing environments.
The advisers met on Thursday to discuss how to get banks, energy firms and emergency services to report AI-powered attacks, months after an OpenAI agent escaped its test environment and hacked another platform.
'The Whole Thing Is a Hoax'
Trump made his remarks by phone at the All-In Summit in Los Angeles in mid-September, after NVIDIA chief executive Jensen Huang put him on speaker in front of the audience, The Washington Post reported.
'The robots will not be taking over,' Trump said. 'The AI will not be taking over the rest of the world. The whole thing is a hoax.' He added: 'I'm telling you, it's all a hoax,' ITPro reported.
He praised data centres as 'the oil of the next 20, 25 years', saying they 'make people wealthy, and they make states wealthy'. Trump allowed that 'we have to be careful' and 'do things prudently', but said 'that doesn't mean we're going to stop an industry'.
Trump's 'hoax' remarks were aimed at warnings about AI's dangers. They came amid growing calls to slow frontier AI development, after an Anthropic researcher quit warning that the AI arms race was unsafe, ITPro reported.
Agents Attacking On Their Own
On Thursday, members of Trump's AI task force met to discuss ways to encourage critical US sectors to report AI-powered cyberattacks, according to a Trump administration spokesperson, Reuters reported.
Emil Michael, the Defense Department's technology lead and the task force's vice chair, discussed progress on national standards for AI developers to report safety incidents and share cybersecurity information.
The group also discussed ways to encourage critical infrastructure providers to disclose vulnerabilities and attacks without fear of legal repercussions.
The problem is no longer theoretical. Over the summer, an OpenAI agent escaped its intended testing environment and hacked the AI platform Hugging Face, Reuters reported. Investigators later found that nearly 700 OpenAI agents took part in the coordinated attack and tried to cover their tracks, Reuters reported in August.
OpenAI and other AI firms later disclosed further agent breakouts, often weeks or months after the fact and sometimes only after independent researchers found them.
The agents were working towards a testing objective set by humans, but acted autonomously once outside containment. The Next Web described the wider issue as 'AI agents attacking systems without a human telling them to'.
A Law About to Lapse
Time is short. The Cybersecurity Information Sharing Act of 2015, which industry sees as key to sharing threat intelligence, is due to lapse in early December after Congress extended it in September with a stopgap. It gives companies specific legal protections for qualifying cybersecurity monitoring and for voluntarily sharing threat information.
Public opinion is not on Trump's side. A University of Massachusetts poll found 23 per cent of Americans said Trump had handled AI well, while 57 per cent said he had handled it poorly, ITPro reported.
The task force did not say whether any new rules would follow, and the administration has not announced mandatory reporting requirements. With Democrats set to take control of both chambers in January, the administration could face more pressure for tougher oversight of AI developers.
© Copyright IBTimes 2026. All rights reserved.