Asked About AI Extinction Risk, the President Said He Worries About China Instead

Asked About AI Extinction Risk, the President Said He Worries About China Instead

A reporter asked President Trump on Thursday whether he had any concerns that artificial intelligence could lead to human extinction. He said he did not. What concerned him, he said, was losing to China.

“No, I don't have any,” Trump told reporters on September 10, according to a Bloomberg report published the next day. “I have concerns that if we don't win AI, we're going to be put in a very bad position.” He then volunteered a number: “We are leading China right now by a pretty good period. I would say a year, which is, you know, considered a lot.”

The exchange lasted seconds. It was also the clearest statement yet of how the administration ranks two competing accounts of what US AI policy is for — one organized around catastrophic risk, the other around geopolitical position.

A Week of Warnings From Inside the Labs

The timing was not incidental. The question reached Trump at the end of a week in which frontier-lab employees broke into open dissent.

On Tuesday, September 8, Anthropic researcher Jacob Coxon announced his resignation on X, writing that the leading labs were “racing straight to self-improving superintelligence and gambling with our lives.” Coxon, who said he spent three years on pretraining research across OpenAI and Anthropic, wrote that people building the technology “earnestly believe it could kill us all by the end of the decade.”

Colleagues who stayed said versions of the same thing. Evan Hubinger, Anthropic's alignment science lead, put his personal estimate of that risk above 10% over the next decade and wrote: “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.” CNBC counted more than a dozen current and former OpenAI and Anthropic insiders publicly calling for a slowdown.

Management responded in kind. In a companywide meeting that week, OpenAI chief executive Sam Altman told employees the company is open to pacing its frontier development and hopes rivals would match it, while acknowledging some would not. OpenAI has separately asked members of Congress whether orchestrating an industry-wide slowdown would survive antitrust scrutiny. A bipartisan July bill, the Collaboration on Adversarial Threats and Security Risks Act, would create a safe harbor for that coordination; it has not been enacted.

Not everyone read the dissent as a safety signal. David Sacks, the administration's AI czar until March, posted that Anthropic's planned IPO should be paused until the researcher's claims could be investigated. He has long argued that American pessimism, not American capability, is the likeliest cause of a lost race.

The Policy Behind the Answer

Trump's remark is consistent with what his government has built. On December 11, 2025, he signed an executive order, Ensuring a National Policy Framework for Artificial Intelligence, aimed at the patchwork of state AI laws. It established an AI Litigation Task Force inside the Justice Department, operational from January 10, 2026, to challenge state AI statutes in federal court as burdens on interstate commerce or as preempted by federal policy. On March 20, the White House released a National Policy Framework built on seven pillars, preemption among them.

There is no comparable federal instrument aimed at catastrophic risk. The proposals that exist are congressional and unpassed: the AI LEAD Act from Senators Dick Durbin and Josh Hawley, creating a federal products-liability cause of action against AI developers; the Ban Artificial Superintelligence Act from Senator Bernie Sanders and Representative Greg Casar; and a duty-of-care proposal circulating among Senate negotiators, reported by Reuters on September 11, that would impose enforceable obligations on developers of the most capable models and could let the government block unsafe releases. Gallup has found 80% of Americans favor safety rules even if they slow the technology.

Analysis: The Race Frame Versus the Safety Frame

These are not merely different emphases; they point in opposite directions on the same questions. The race frame treats capability lead as the safety-relevant variable: a slowdown hands the frontier to a government with fewer constraints, so speed is the mitigation. The safety frame treats the lead as beside the point — if the failure mode is loss of control, it matters little which flag flies over the lab that loses it. Each is internally coherent, and each implies a different answer on preemption, on liability, and on whether an antitrust safe harbor for coordinated pacing is a public good or a cartel.

The Senate duty-of-care draft is where the collision becomes concrete. It would be the first statutory acknowledgment that frontier risk is a regulatory category — while sitting atop a preemption architecture the executive branch is already litigating to expand, meaning one bill could create a federal obligation while extinguishing state ones. Polling complicates that half: a YouGov survey for the Institute for Family Studies found 57% of Americans opposed to congressional efforts to block state AI regulation against 19% in support.

Then there is the number itself. A one-year lead is contested rather than measured. DeepSeek's own technical paper for V4 put the model three to six months behind state-of-the-art frontier systems. The Center for AI Standards and Innovation found roughly an eight-month gap to DeepSeek V4 Pro; by March 2026, Anthropic's Claude Opus 4.6 led the strongest Chinese model by about 2.7 percentage points. On inputs the gap is far wider — roughly 10x in installed compute and 35x in advanced chip manufacturing. Twelve months is defensible as an upper bound on the capability gap and an understatement of the hardware gap. It is not a figure any single source produces.

What to Watch

Whether the Senate duty-of-care text arrives standalone or attached to must-pass legislation, and whether it preserves or preempts state authority. Whether Congress moves on an antitrust safe harbor, without which Altman's coordinated-pacing idea has no legal path. And whether the DOJ task force files against a major state statute this fall — the first court test of an approach the President's answer on Thursday explained more plainly than any executive order has.