The AI slowdown its leaders are calling for could also lock in their lead over the rest of the sector

By Zendoric · September 24, 2026.

In recent months, the leadership of the artificial intelligence industry has converged on a surprising message: it is time to slow down. Dario Amodei (Anthropic), Sam Altman (OpenAI), Elon Musk (SpaceXAI), Demis Hassabis (Google DeepMind) and Satya Nadella (Microsoft) have each, in their own way, voiced support for slowing the development of the most advanced AI, according to NPR. None of them shares the same plan for achieving it, but the underlying consensus—that the current pace is risky—is unprecedented among rivals who until recently competed with no visible brakes.

The problem, their critics say, is who comes out ahead if that brake is actually applied. "If there's an industry-wide freeze, essentially what will happen is that whoever has the strongest model today will dominate market share," sums up David Bellamy, a researcher at the Institute of Foundation Models, an open-source AI center backed by the United Arab Emirates. Bellamy estimates that his institute's model is about six months behind the most advanced ones from OpenAI and Anthropic; freezing that gap, he argues, would be especially damaging just as both companies prepare to go public.

The criticism does not come only from small labs. Alvaro Bedoya, a former commissioner of the Federal Trade Commission (FTC) during the Biden Administration, warns that a pause could tilt the playing field: "rivals, new entrants, the smaller players in that market can no longer compete on product, quality or service. They're simply shut out. And the options left to them are to be bought by the incumbent or to disappear." Perplexity responded to NPR with a line that captures the skepticism: one should be wary "any time a company asks to be regulated in a way that protects its market position." And Nick Frosst, co-founder of Canada's Cohere, went further: he called a "privately coordinated pause" a measure that "is not a neutral safety policy," but rather a mechanism that would lock in the advantage of the small group of labs that already concentrates the most compute, capital and distribution, shutting out researchers, startups and other labs.

The disagreement has already reached the courts. Four consumers of AI products filed a federal lawsuit last week against Anthropic, OpenAI, Google and SpaceXAI, accusing them of reaching an illegal agreement to slow the pace of AI development. According to the suit—whose accusations are, for now, unproven allegations—that pact would violate antitrust law and slow improvements to the paid subscriptions consumers use.

Hanging over all of this is the question of China. Amodei himself acknowledged in a recent essay calling on the industry to "pace the frontier" that any negotiated pause would require the cooperation of Beijing, "the autocratic country that, by a wide margin, has the most advanced AI capabilities," and that any such agreement "must have ironclad verifiability, or be limited enough that breaking it would not be militarily existential." Susan Rice, former national security adviser under Obama, argued in a New York Times opinion piece for a pause negotiated with China until both countries can guarantee that no frontier AI escapes human control. Trump, by contrast, dismissed fears about runaway AI on Truth Social as overblown and insisted that "the only one who is happy about this is China"—a stance that takes on particular relevance this week, with Xi Jinping meeting Trump in Washington and AI on the agenda. No one has yet explained how Chinese compliance with such a pact would be verified; some observers quoted by NPR compare a potential agreement to a nuclear non-proliferation treaty, an analogy that is no accident: it implies inspections, mutual trust and credible sanctions that do not exist today between Washington and Beijing on AI.

A note on transparency worth flagging: Anthropic is one of NPR's funders, although the outlet clarifies that it does not sell coverage or interviews in exchange. That does not invalidate the reporting's facts, but it is worth keeping in mind when reading Amodei's quotes in a debate where his company has direct commercial interests.

Our take: this is exactly the kind of short-term risk we have been flagging for months in this editorial line—the concentration of power—only here it comes dressed up as a gesture of responsibility. We do not think Amodei's concern about loss of control is feigned: Anthropic has for some time been publishing, on its own initiative, internal evaluations that acknowledge blind spots in overseeing its models' reasoning, and that counts for more than any statement of intent. But genuine concern about risk is one thing, and the specific remedy being proposed is quite another. A "voluntary" pause among a handful of labs that already concentrate compute, capital and distribution is not neutral: by design, it freezes the current ranking at the worst possible moment for anyone trying to close a six-month gap, and at the best possible moment for anyone preparing to go public with that advantage intact.

This connects to something we have argued here before: the democratizing force of open-source AI is one of the sector's best pieces of underlying news, because it distributes capability instead of concentrating it in a few hands. If the brake on the frontier is applied only among those already at the top, the result is not necessarily less systemic risk: it is less competition. And without competition, the promise of abundance that underpins our underlying optimism—an AI that spreads progress rather than hoarding it in a few hands—becomes harder to deliver, not easier.

The piece that really decides whether this debate is about safety or about market share is China. Without a credible verification mechanism with Beijing, any pause among American companies is, in practice, a competitive advantage for the leaders over their domestic rivals, not a real barrier against the biggest geopolitical rival. This week's meeting between Trump and Xi, and the outcome of the antitrust lawsuit against the big four, will tell us whether "pacing" remains a corporate narrative or becomes real governance, with verifiable mechanisms and, hopefully, with room for those who are not among the five names now calling for a slowdown.