Dario Amodei asked AI labs to slow down. Anthropic's IPO filing says that call isn't about its own release timelines.

Anthropic's public IPO filing draws a distinction between Dario Amodei's call to slow advances in AI and the company's own plans. Its appeal for "pacing the frontier," the filing says, "does not describe our own model development or release timelines."

That qualification follows Amodei's September 12 essay, in which Anthropic's chief executive wrote that "we must slow the pace at which we improve the capabilities of AI models" and warned that "a race to the bottom, spurred by commercial incentives, can make these risks more acute." The same day, Sam Altman said OpenAI would match Anthropic's commitment to outside evaluators and told Fortune that this was "an ill-advised moment to go public."

Anthropic made its filing public on October 26 without disclosing an offering size. The filing also describes steps Anthropic says it is taking on safety. It recently delayed pre-training a new frontier base model for more than a month to tighten security around its model weights and intellectual property, and it expects to invest at least $1 billion over the next five years "to build external reviewer capacity with Accenture." The money is part of a program in which independent evaluators work alongside Anthropic's risk and safety teams "with broad internal access," and the terms with Accenture "have not been finalized and are subject to change."

What the S-1 says about safety and pace

In its business section, Anthropic defines "pacing the frontier" as industry-wide standards "primarily related to the use of AI to accelerate AI development," so that capabilities do not advance faster than safety, alignment, and security practices. The company's own timelines "continue to be informed by our risk assessments," the filing says. If standards like these are adopted industry-wide through government action or regulation, the rate at which AI capabilities advance "could be faster or slower than it is today."

In its risk factors, under the heading "Balancing Safety and Commercial Objectives," Anthropic says its commitment to safety "requires us to make difficult judgments that balance safety considerations against commercial objectives, competitive pressures, and stakeholder expectations." The company says it may delay or pause a stage of training "when we are not confident in our safety measures for that stage."

Concerns have been raised inside Anthropic, and by outside researchers, regulators, policymakers, and industry participants, that sufficiently advanced AI systems could "pose catastrophic or existential risks to humanity," the S-1 says. Anthropic writes that its models "are contributing to the pace of our own research," and that those risks will increase as AI systems automate AI research and development "if, for example, they accelerate AI progress beyond our ability to evaluate and safeguard new models."

In a passage on securing computing power, the S-1 says "the pace of AI development is extremely rapid," and "even a brief period without access to sufficient compute could cause us to fall behind quickly."

September's IPO reports

The essay appeared amid a series of reports about Anthropic's IPO preparations. Reuters reported on September 4 that Anthropic expected to begin marketing the offering in mid-October and to list before the November 3 midterm elections. On September 11, Reuters reported that Anthropic was seeking to raise as much as $100 billion at a valuation near $2 trillion, and that Nvidia (NVDA) was in talks to anchor the offering with as much as $10 billion. Anthropic hasn't commented on the anchor talks. The timetable has since moved. Bloomberg reported on October 1 that Anthropic is aiming to go public as soon as mid-November, and Reuters reported on October 15 that some investors at Anthropic's October 14 investor day discussed a valuation of around $1.5 trillion.

What Amodei asked for, and what he did not say

The essay's one unilateral commitment is to embed third-party evaluators inside Anthropic with desks, badges, laptops, and the right to publish findings "without editorial control by Anthropic." The rest depends on other companies and on governments: shared safety standards among labs in democratic countries, which he writes would need the US government to mediate or grant an antitrust waiver, and then a four-level arrangement with China. Pacing doesn't mean halting training, he writes. It means time spent on monitoring, alignment, interpretability, and evaluation. The essay does not mention the IPO, investors, shareholders, or Anthropic's valuation.

He gives no figure for how much slower. The essay says interpretability could make "profound progress in 1-2 years," and that a misaligned agent swarm could within "6-12 months" be capable of taking over the internet. It doesn't say what a slower Anthropic would spend less on, or whether a company reportedly seeking to raise up to $100 billion would buy less computing capacity. Anthropic's compute agreements could reach $517 billion over about a decade, The Information estimated on September 6, a figure the company hasn't confirmed.

Before the S-1, the company had addressed the question in a policy post titled "When AI builds itself." Anthropic staff Marina Favaro and Jack Clark wrote that the company would slow down if other frontier developers did so "in a verifiable manner." It is conditional, and Anthropic has not said that it considers the condition met.

The people who have connected safety and the listing

Altman is the most prominent executive to connect safety and listing timing on the record. Asked by Fortune whether 2026 was off the table, he said, "I would say not 2026. Yeah, we got a lot of stuff to do, like meeting this moment of what is going to be required for safety and alignment." OpenAI filed its own confidential S-1 in June, and its chief financial officer told employees in August the company would be public in 2027.

On September 8, Jacob Coxon, a pretraining researcher, resigned from Anthropic and wrote on X that "neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives." He left two months before his Anthropic stock vested, he told Axios, adding, "I no longer have anything to gain by juicing up Anthropic's valuation."

Later that evening, Evan Hubinger, who leads alignment science at Anthropic, quoted a post in which Coxon wrote that the people building AI "earnestly believe that it could kill us all by the end of the decade." Hubinger wrote, "Jacob is correct here," and added, "we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade." He also 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."

Anthropic has since released two models in six days. We looked at what its pacing commitment would actually hold back.