Trump calls AI risk overstated on 60 Minutes, then describes a jail threat to an unnamed lab chief

On 60 Minutes, asked by Ross Douthat about AI executives’ public warnings over their own models, President Trump said the dangers are smaller than claimed and the risk is already being handled. His example was striking: the Justice Department called the head of one of the big AI companies, after a report of something bad done with AI, and told that person to get ready for jail. He wouldn’t say who, or what the conduct was. He also said he would work with Democrats on guardrails, but not at the cost of slipping behind China, and floated calling AI “SI,” for super intelligence, echoing the administration’s new terminology (super intelligence).

If accurate, the strongest lever over frontier AI in the US may be private, unexplained pressure from the top, not published rules.

Two things should be separated. On substance, reasonable people disagree about how near or severe the worst AI failures are, and calling the danger manageable is defensible; the case for letting useful systems ship and fixing problems as they appear is real. WIRED’s experts point to concrete risks rather than extinction (10 experts).

The method is the problem. Managing risk by phoning an executive and threatening prison, with no named company, stated offense or public process, is the opposite of predictable oversight. Trump says the person changed everything afterward; there is no way to check what changed, whether it made anything safer, or whether the original report was true. The administration already describes but won’t publish its safety-testing framework (secret rules).

Scaled up, labs learn the real compliance department is political proximity. Favored companies get latitude; others get calls. Safety becomes whatever the person on the phone wants that week, invisible to the public. That is bad for safety advocates and for builders who want stable rules, and it contrasts with explicit proposals such as Anthropic’s threshold bill (Anthropic bill) or Scott Wiener’s federal agenda (Wiener).

Trump said China has no guardrails except whatever Xi says, then remarked that his own control is strong but not quite that strong, framing centralized authority as the benchmark. The US is better off competing on transparent rules markets and courts can see. The offer of bipartisan guardrails is the most hopeful line; the test is whether it becomes anything written down. More in our regulation coverage and AI coverage.