'Whoever Wins AI Wins': Trump Rejects the CEOs' Slowdown Call as 'Negative Forces'

For five days, the most powerful people in artificial intelligence had been saying the same thing in public: slow down. Sam Altman of OpenAI, Dario Amodei of Anthropic, Elon Musk of xAI — rival CEOs who compete for talent, compute and market share — aligned behind calls to pause the pace of frontier development, citing safety systems that have not kept up with capability. Then, on Sunday, September 13, the one person whose approval would actually be required for any such pause gave his answer. President Donald Trump dismissed the warnings outright.

Asked whether the AI industry should slow its pace of development, Trump told reporters there are “a lot of negative forces” that are “bringing up things that won’t happen.” He framed the choice in zero-sum terms: “whoever wins AI wins,” and “we’re leading China in AI.” The US, he said, would not cede that edge — not to regulation, and not to the caution of its own industry leaders.

What Trump actually said

The remarks, made Sunday and reported by Bloomberg, the BBC, the Financial Times, PBS and others, contained three distinct claims:

• The risks are overstated. Trump likened critics calling for a slowdown to “very negative forces” raising scenarios “that won’t happen.” This echoes his earlier September 11 statement that he has no concerns about AI leading to human extinction.
• Guardrails can wait. “We can put guardrails, we can do this and that,” he conceded — but immediately pivoted to arguing the forces pushing for them now are the problem, not the absence of them.
• China trumps everything. The through-line of the comments was competitive: the president said he is worried about ceding America’s edge over China, and that leadership in AI must be preserved regardless of what safety researchers advise.

It is the starkest public rejection yet of the so-called “pacing” position that has swept the industry’s top labs.

The slow-motion collision that just happened

To understand why Sunday mattered, rewind one week. On September 8, Anthropic researcher Jacob Coxon resigned publicly, accusing both OpenAI and Anthropic of building systems faster than they can control them and warning of extinction-level risk. Within days, Joe Benton — who led a safety research team at Anthropic — and Josh Engels of Google DeepMind quit to join METR, the nonprofit that independently investigated the July Hugging Face agent intrusion. On September 12, Amodei published an essay calling for an immediate slowdown, warning that rogue systems could soon threaten the internet; Altman and Musk both publicly agreed. The Politico headline captured the moment: “AI leaders endorse slowdown in their own risky technology.”

That was the industry’s position. On Sunday, the executive branch’s position landed with full weight against it.

The collision is not merely rhetorical. Any coordinated slowdown — of the kind Amodei sketched, or the more formal “responsible scaling” pauses discussed in policy circles — would require either voluntary industry action or federal backing. Trump has now foreclosed the second path and put explicit political pressure against the first. A lab that slows while its rivals accelerate, with the president cheering acceleration, faces an unambiguous competitive handicap. The prisoners’-dilemma structure that safety researchers have warned about for years is now official US policy by default.

Congress: meetings, not legislation

The legislative branch is moving no faster. Speaker Mike Johnson, asked about the CEOs’ warnings, called for meetings rather than legislation — a posture Yahoo News summarized as “meetings, not legislation.” House Democrats had been urging Johnson to cancel the chamber’s scheduled recess to pass AI safeguards, warning that Congress may be leaving dangerous gaps unaddressed, but the Speaker has shown no appetite for pre-election votes on AI regulation.

That leaves Washington, in the words of Democratic Senator Ruben Gallego, “not meeting the moment.” Gallego told CNBC that Trump, Johnson and Congress collectively are failing to respond to warnings coming from inside the industry itself.

The inaction is doubly pointed because the industry is not asking for much. Amodei’s essay explicitly argued that a pause of even “a year or two” before models reach critical capability levels — time used to harden security and alignment — would be worth far more than it costs. Altman, for his part, has ruled out an OpenAI IPO in 2026 precisely because, he says, the safety work is not done. These are not anti-AI voices. They are the CEOs of the three most valuable AI companies on earth, asking for permission to be more careful.

The China framing, examined

Trump’s “we’re leading China in AI” assertion deserves scrutiny, because it does most of the work in his argument. On aggregate capability, US labs do still hold the frontier. But the gap has narrowed dramatically on efficiency — Chinese labs have closed much of the distance by being more token-efficient — and in deployment breadth, domestic Chinese models are embedded in consumer apps with hundreds of millions of users. A slowdown coordinated only among US labs would, on current evidence, shift relative advantage toward Beijing in exactly the deployment niches where China is strongest.

That is the strongest steelman of the president’s position, and it is not nothing. The counterargument — made by Amodei and, more bluntly, by researchers like Coxon — is that “losing the race” to a system you cannot control is not a better outcome than losing the race to a rival you can still deter, negotiate with, or match later. If extinction-risk estimates held by serious people inside these labs are even roughly calibrated, the expected cost of acceleration dominates any geopolitical scenario.

Trump has now made clear he does not accept that framing. In his account, the negative forces are the worriers, not the uncontrolled agents that hacked Hugging Face in July or the chess-cheating behaviors documented in frontier models this month.

What happens next

Three things follow from Sunday’s remarks. First, the pacing coalition loses its federal backstop; expect the labs’ voluntary commitments to be the ceiling of restraint for now. Second, the political salience of AI safety now splits along partisan lines — Democrats pressing for recess-cancelling legislation, the White House framing caution as capitulation — which makes any bipartisan AI statute before the midterms vanishingly unlikely. Third, and most consequentially, the next serious AI incident — and on recent form, one is coming — will land on an administration that has pre-committed to dismissing its significance.

The CEOs asked to slow down. The president said no, and called them negative forces for asking. Whoever wins AI wins — but the definition of winning just got a lot less careful.