Asked Whether AI Could Kill Everyone, Trump Said His Only Worry Is Losing to China

On the flight back from the Dallas convention on September 10, a reporter asked Trump whether he had heard the warnings that artificial intelligence could destroy humanity.

"Not worried at all," he said. "What I worry about is that if we don't win the AI race, we'll be at a very big disadvantage. We're ahead of China by a pretty big margin — about a year. And that's a big gap."

That is the entire risk assessment of the US government, delivered in three sentences: extinction risk, zero; China gap, one year; policy, run faster.

Who He Was Answering

The question came from inside the industry, not from its critics.

On September 8, a safety researcher leaving Anthropic wrote publicly that "AI companies are putting human lives at risk" and that "the people building AI genuinely believe it could kill us all within ten years" — telling the WSJ that at the current pace, systems could be effectively out of control "by the end of next year." A current Anthropic researcher, Evan Hubinger, responded that he personally puts the probability above 10% within a decade.

Two days earlier, OpenAI's chief scientist Jakub Pachocki had posted to an internal company board that "no one seems prepared for the consequences of the continued rapid advancement of machine intelligence." His description of why is the technically important part: modern AI systems are not designed — they grow, through optimization repeated at unimaginable computational scale, and "their overall workings are beyond our ability to understand." His conclusion: slow down until safeguards are certain, and coordinate internationally between governments and industry.

Note what Pachocki asked for, because it defines what Trump refused. "Slow down until safeguards are certain" is only possible as a coordinated act — a lab that slows alone loses to the labs that don't, and a country that slows alone loses to China. The only actor who could impose coordination on US labs is the federal government, and the only mechanism for extending it to Beijing is the kind of international negotiation Pachocki named. Trump's answer forecloses both. "We're a year ahead" is not a rebuttal of the risk argument; it is a statement that the race frame outranks it — and in a race frame, every safety pause is a gift to the adversary. The researchers understand this, which is why the warnings are getting louder as they get less actionable.

The China Answer, Every Time

This is now a documented pattern rather than an inference. When communities resisted data centers over electricity and water, Trump's answer was that "China would love nothing more" than the opposition. When export controls and AI competition policy come up, the frame is the same. Now extinction-risk warnings from the labs' own researchers receive the identical response.

Every domestic objection to the AI buildout — local, environmental, or existential — is being routed to the same answer: we cannot afford it, because China. The claimed one-year lead functions as a permanent emergency: large enough to protect ("a big gap"), small enough that any slowdown forfeits it.

What It Means for Markets

The near-term read is unambiguous and bullish for the buildout. A president who dismisses the strongest possible objection to AI acceleration — raised by the builders themselves — will not impose compute caps, safety-review gates, or licensing regimes that bind. US AI capex has a policy floor under it: chips, data centers, power. The regulatory-risk discount that periodically surfaces in AI-exposed names should stay compressed while this administration governs.

The honest asymmetry is in the tail. Markets have no instrument for pricing "the people building the product publicly estimate double-digit odds it kills everyone within a decade," so they price it at zero — the same way the president does. What is priceable is the nearer step: if a frontier system produces a visible, attributable failure — financial, infrastructural, biological — the swing from "zero regulation" to "emergency regulation" would be violent precisely because nothing was priced in between. Insider warnings of this volume and specificity are how that scenario looks in advance.