Trump Rejects AI Slowdown Calls — 'Whoever Wins AI, Wins'

Twenty-four hours after the frontier's leading CEO asked Washington for permission to slow AI down, Washington answered from a golf course on Ireland's west coast. Taking questions at his Doonbeg resort on Sunday, President Trump was asked whether the AI industry should slow down or face more regulation. His answer, per Reuters: "We're leading China in AI. We're the most sophisticated country in the world, and frankly I want to keep it that way because whoever wins AI wins." The extinction scenarios driving this week's slowdown calls? The work of "very negative forces... bringing up things that won't happen."

What Actually Happened

Per Reuters' dispatch from Doonbeg, the question came as the pacing debate peaked. On Saturday, Anthropic CEO Dario Amodei published a three-step framework to pace frontier development — embedded third-party evaluators, coordinated pacing among democratic-country labs, eventually verified global coordination — and the proposal drew immediate endorsements from other major AI executives, including Elon Musk and OpenAI CEO Sam Altman. We covered the plan and its fine print yesterday; the mechanism it turned on was a request for a narrow US antitrust waiver, since rivals agreeing to slow their own progress is otherwise a textbook coordination problem.

Sunday morning, the plan's author was already conceding its hardest part. On CBS's "Face the Nation," Amodei called US-China competition the "toughest dilemma" in establishing a speed limit on AI development. "We shouldn't kid ourselves, because the incentives to pull ahead and the military advantage that you get from that are so large that the ability to check that the other side isn't cheating has to be ironclad," he said. "So I think that's going to be the work of years, and honestly, I don't know if it's possible." Jacob Coxon — one of the Anthropic researchers who resigned this week and publicly warned about extinction risk — told NBC's "Meet the Press" the US needs international coordination, "or else we risk running the same race with China, which could be equally dangerous."

Then Trump closed the loop. Reuters notes he has consistently resisted stirring concerns about AI; on Friday, asked about existential risk by Politico, his answer was "I don't have any." Sunday's remarks extended that from indifference to an affirmative case: the race is the policy. "We could put guardrails. We can do this and that," he said — a caveat worth keeping — "But I think you have a lot of very negative forces that are bringing it up that shouldn't be bringing it up, and they're bringing up things that won't happen." The debate Trump is dismissing has a specific origin. Two Anthropic researchers' warnings — that rapidly progressing AI could lead to human extinction "in the not-too-distant future," as Reuters summarizes it — went viral this week, and Coxon's resignation letter landed days before his boss's essay. Reuters also frames the political economy around the question Trump was asked: his administration has largely embraced AI firms in its second term, those firms have supported the administration's initiatives, and AI executives are pouring money into the midterm campaign ahead of the Nov 3 election that will determine control of Congress.

The Waiver Died in Doonbeg

Yesterday's analysis flagged the antitrust waiver as the single instrument the entire pacing architecture turns on: step two — democratic labs coordinating on pacing standards — is only legal if the government enables it. The person who issues that waiver was just asked, directly, whether the industry should slow down, and he called the plan's premise a thing that won't happen.

Without the waiver, step two degrades into what Amodei's own essay implied it would be without government backing: a legal exposure for any lab that participates. That was the reason he asked for government mediation in the first place. With a White House that has "largely embraced AI firms," per Reuters, and that reads the current alarm as the work of negative forces rather than a policy input, no waiver is coming — which means no coordination mechanism, which means the plan's binding steps reduce to what Anthropic can do alone: embedded third-party evaluators with publication rights. A real commitment, but a company policy, not a market structure.

The counterpoint inside Trump's own orbit underscores the ceiling. The president's former AI czar told Anthropic and OpenAI to slow down on Sunday — and to stop asking for anything in return, per The Next Web. Even the sympathetic version of the slowdown argument in this administration is: pace yourselves voluntarily, no waivers, no immunity. That is not what the pacing essay needs to function.

The Verification Problem Was Never Political

Amodei's Face the Nation quote deserves its own read, because it's the most honest sentence of the weekend: the binding constraint on global pacing isn't Washington's willingness — it's verification. An agreement the other side can cheat on hands the defector the military advantage, so it must be "ironclad," and the CEO proposing the regime says he doesn't know if that's possible. If the chief architect rates the core mechanism at "unknown," the plan was always going to live or die on trust-based voluntary measures — precisely the ones antitrust law polices.

Strip out the politics and this was the predictable endgame. The plan asked for a level of statecraft — verified, enforced coordination between rivals and eventually between superpowers — that no current institution provides, in a timeframe measured in months. Trump's answer didn't kill the pacing idea so much as reveal that it never had a buyer.

The Only Policy Is the Race

What Washington is actually doing breaks along the campaign calendar. The Nov 3 midterms will decide control of Congress, AI executives have poured money into the campaign, and Musk's America PAC has already spent millions backing Republican candidates, per Reuters. Democrats are increasingly attacking the administration's hands-on-for-industry, hands-off-for-regulators approach as the election nears (Washington Post). Hence Sunday's chorus: House Minority Leader Hakeem Jeffries called on lawmakers to "act urgently" on AI (ABC News); Speaker Mike Johnson said tech companies must be primarily responsible for AI safety and that AI leaders must meet with Congress following the "end of humanity" warnings (Politico, New York Post); Senator Bernie Sanders proposed up to 20 years in prison for developers who plow ahead with artificial-superintelligence plans, a penalty he'd put on par with illegally developing rogue nuclear weapons (Tom's Hardware). The New York Times' Sunday assessment of all this: Washington stirs, but mostly slumbers.

Even the government's own buyer's-side experience cuts against a slowdown. Reuters notes the Pentagon warned of risks from Anthropic's model this year — the safety-leading lab's product, flagged by the single largest AI buyer in the federal government — and the policy response remains acceleration.

And the other side of the Pacific isn't waiting for anyone's speed limit. Xi Jinping told BRICS counterparts Sunday that China will lead AI and tech cooperation among the bloc's countries (CNBC), and Z.AI's roughly $5 billion Hong Kong raise — stacked on $4 billion in July — is already earmarked for the compute and talent to compete with US rivals. The two-speed market we flagged yesterday now has its answer on the American side: the throttle isn't coming from this White House, and it was never coming from Beijing.

What Builders Should Take From It

• Price for continuous churn, not a speed limit. The pacing regime had exactly one possible buyer — the US government — and it passed. Model jumps, price cuts, and capability churn continue at full pace; model-agnostic architecture and quarterly vendor re-quotes stay mandatory.
• Treat voluntary safety as marketing until it's verified. Embedded evaluators and risk essays are company commitments with no regulatory teeth. Weight actual behavior — pricing stability, incident disclosure, uptime — over safety positioning when you choose vendors.
• Watch November, not Congress. With seven weeks to the midterms, expect a loud AI-regulation debate and near-zero statute. Procure for what exists today and revisit if control of Congress flips.
• Your agent controls are still the only controls. Amodei's 6–12 month window before a misaligned agent swarm could run internet-scale damage doesn't get a federal program out of this White House. Sandboxing, egress allow-lists, grader isolation, and kill switches are your responsibility — ship them as product requirements, not ops afterthoughts.
• "We could put guardrails" is the door left open. If anything moves in Washington this year, expect it to look like the administration's OpenAI partnership that secured federal agencies a 50% discount on frontier models (Fox Business) — pro-adoption guardrails, not pacing.

The signal from Doonbeg isn't that Trump settled the AI safety debate. It's that he declined to buy the industry's ask. A pacing regime needs a sovereign to underwrite it, and the sovereign who was asked spent the weekend explaining why winning the race matters more than slowing it.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.