Amodei’s AI Slowdown Call Drew Altman’s Support—but Triggered a Market and Political Clash

Anthropic CEO Dario Amodei’s appeal to slow the improvement of the most capable AI models quickly became more than a safety argument. It produced rare public alignment among leading AI executives, pushed OpenAI’s anticipated public listing beyond 2026, rattled AI-exposed equities and drew a blunt rejection from President Donald Trump. 1710

What Amodei proposed

Amodei did not call for ending AI research. His proposal was to “pace the frontier”: deliberately moderate the speed at which frontier labs improve model capabilities so that safety work, oversight and coordination have a chance to catch up. He argued that increasingly capable systems could become harder for people to understand and control, particularly if AI materially helps build the next generation of AI. 177

The plan had three broad elements:

1. Independent oversight inside frontier labs. Anthropic said it would provide third-party evaluators with permanent, employee-level access to its systems. Those evaluators would be able to verify adherence to safety measures, report incidents and assess model alignment during training. 17
2. Coordination among frontier AI companies. Amodei called for shared safety standards and limits intended to prevent an unchecked competitive race. 17
3. International cooperation. The proposal envisioned cooperation across governments to manage the risks of advanced AI. 17

That first measure matters because it shifts part of safety assessment outside a company’s own internal review process. It is a commitment to ongoing access and scrutiny, not simply a one-time external audit.

Sam Altman backed the call—and postponed OpenAI’s IPO

OpenAI CEO Sam Altman publicly endorsed Amodei’s position, writing that OpenAI also needed to “pace the frontier.” Reports described the response as a notable show of agreement between leading competitors in the race to build increasingly powerful models. 1

Altman paired that support with a consequential business decision: OpenAI would not go public in 2026. He described an IPO at that moment as “ill-advised,” saying the company had substantial safety and alignment work to address. Reuters reported that Altman regarded even a 10% chance of AI-caused human extinction by the end of the decade as unacceptable.

The practical implication was clear: OpenAI was presenting safety work as a priority that could outweigh the timing of one of the technology sector’s most anticipated listings. It did not amount to a detailed public timetable for slowing models, but it put the company on record supporting a more cautious pace.

The researcher warnings that set the stage

The call followed the resignation of Jacob Coxon, a researcher who said he had worked on pretraining research at both OpenAI and Anthropic. Coxon accused the companies of “gambling with our lives” in a race toward self-improving AI systems. 18

The discussion also included comments from Anthropic Alignment Science Lead Evan Hubinger. According to CNBC’s reporting, Hubinger said he personally put the chance that AI could kill all humans this decade above 10%. That was an individual risk assessment reported in the public debate—not an Anthropic corporate probability estimate or a demonstrated forecast.

These warnings helped turn a long-running, often abstract dispute about catastrophic AI risk into an immediate question for company leaders: whether labs should continue maximising capability gains while safeguards are still being developed.

Why markets sold AI-linked stocks

Investors interpreted the slowdown call as a potential challenge to assumptions behind the AI trade: sustained increases in model capability, heavy spending on compute and strong demand for chips and supporting infrastructure. Asian AI-linked stocks dropped after the announcements, and U.S. equity futures also declined.

SoftBank, a major OpenAI investor, fell as much as 13.2% in Tokyo, according to Reuters. The same report said the reaction spread across AI-related shares in Asia. Other reporting put Nasdaq futures down about 1.5% in premarket trading as chipmakers came under pressure.

A market move does not prove that AI investment or chip demand will collapse. Rather, it showed that investors were repricing a risk: if frontier-model advances are deliberately constrained or delayed, expected revenues and returns across model developers, infrastructure providers and AI investment vehicles could arrive later than previously assumed.

Trump rejected the slowdown argument

Trump rejected the call to slow development, framing AI primarily as a contest for U.S. strategic leadership over China. “Whoever wins AI wins,” he said, while arguing that the United States was ahead and should stay ahead. 10

That response captured the central political conflict. Safety advocates argue that more time for testing, alignment research and independent oversight can reduce the chance of severe harm. Opponents of a slowdown fear that unilateral restraint could hand an advantage to states or companies that continue advancing capability without comparable limits. 1710

Amodei acknowledged that international non-cooperation—especially from adversarial countries such as China—was the hardest problem for a pacing regime. A policy that applies only to some frontier labs could make safety-oriented firms bear commercial and geopolitical costs while leaving the underlying capability race intact.

What the dispute means for AI companies and public markets

The episode left an unusual mix of consensus and conflict:

• Lab leaders found common ground on the need for stronger safety measures. Altman supported Amodei’s call, while Anthropic committed to embedded third-party evaluation.
• The operational details remain unresolved. Public support for “pacing” does not by itself establish common thresholds, enforcement mechanisms or a binding international agreement. 17
• OpenAI’s IPO was directly affected. Altman ruled out a 2026 listing, citing safety priorities.
• Anthropic’s public-listing path was less clear from the available reporting. The sources support its evaluator commitment and broader policy proposal, but do not establish that Anthropic formally changed IPO timing because of the dispute.

Ultimately, the argument is not simply about whether AI should advance. It is about who sets the pace, what independent evidence is required before models become more capable, and whether safety commitments can survive the commercial and geopolitical pressure to move faster. 1710