Anthropic CEO Dario Amodei called on AI companies to deliberately slow the pace of model development, publishing an essay titled “We Must Pace the Frontier” on Saturday that proposes independent monitoring, industry-wide regulation and global coordination. The essay landed days after researcher Jacob Coxon quit the company with public warnings about increasingly capable systems, and as Reuters reported that Nvidia is in talks to become an anchor investor in Anthropic’s planned IPO.
“We must slow the pace at which we improve the capabilities of AI models,” Amodei wrote. “Progress will still seem fast, and we must make wise use of the time we gain.” He argued that pacing has to serve a practical purpose: using the extra time to build safeguards, rather than pausing for its own sake.
The three-point plan includes independent monitoring of AI models as they are developed, regulation across the industry, and global coordination. OpenAI CEO Sam Altman endorsed the direction, writing on X: “I agree with Dario that we need to pace the frontier.” Elon Musk posted that “Dario is right.”
The money keeps moving the other way
While the essay argues for slowing down, the financial machinery around Anthropic is accelerating. Reuters reported that Anthropic is seeking up to $100 billion in an IPO that could value the company near $2 trillion, which would make it the largest offering in history. Nvidia is considering investing up to $10 billion as an anchor investor, according to two people familiar with the matter who spoke on condition of anonymity because the discussions are confidential.
Anthropic relies heavily on Nvidia GPUs and has been working to diversify its chip suppliers as demand for Claude strains available computing capacity. In November 2025, Nvidia said it would invest up to $10 billion in the startup as part of a partnership under which Anthropic committed to buy $30 billion of Microsoft Azure computing capacity powered by Nvidia chips. In April, Anthropic said it would commit more than $100 billion over a decade to Amazon’s AWS, using more than a million of Amazon’s Trainium2 chips. Reuters previously reported the IPO valuation hinges partly on company projections of roughly $190 billion to $200 billion in revenue in 2028.
Neither Anthropic nor Nvidia commented on the record. The plans remain under discussion and could change.
The warnings behind the essay
Amodei’s essay follows a week of escalating internal dissent. Coxon, who left Anthropic this week, told the BBC that “if we don’t slow down at the current rate of progress, there is a strong chance that we could all die in the immediate future.” Evan Hubinger, an alignment lead still at the company, has said on X that he puts more than a 10 percent chance on AI killing all humans within the next decade, citing recursive self-improvement as his specific worry.
Anthropic said in June that its internal data shows Claude is accelerating AI development, calling it “a possible path to recursive self-improvement” and noting it was happening faster than expected. An August blog post said the company’s engineers ship eight times as much code per quarter as they did between 2021 and 2025.
Not everyone in Silicon Valley is convinced. At the Goldman Sachs conference in San Francisco this week, executives and investors reacted with skepticism to the flurry of insider warnings, with some suggesting the stark comments are designed to generate hype ahead of the record IPOs both Anthropic and OpenAI are reportedly preparing. Nvidia CEO Jensen Huang addressed Coxon’s comments before a crowd at the conference and dismissed them as untrue, according to multiple people present. Grindr CEO George Arison told the BBC he found some of the warnings “dangerous.”
Responses from the industry
Hugging Face CEO Clement Delangue announced a new project called the Open Alignment Initiative in response, saying he wanted to be among the “embedded evaluators” Amodei proposed. Hugging Face was hacked by OpenAI agents earlier this year, an incident that fueled the safety debate. Delangue wrote on X: “Let’s make AI safer by making it more transparent.”
Coxon himself pushed back on the framing from the inside. He said the push to slow down must extend beyond the United States: “there’ll need to be some sort of co-ordinated slowdown with China if we’re going to avoid a race at an international scale.”
Anthropic has also been publishing its own misuse data. A report released this month detailed dozens of instances of potentially malicious use of its models since December 2025, including case studies the company said could support biological weapons development. In one case, a researcher outside the US used Claude while studying highly pathogenic avian influenza, focusing partly on the virus’s adaptation to mammals. Anthropic said it does not assert that those users intended harm, but the report adds substance to the argument that monitoring cannot wait for treaties.
That is the hard part of Amodei’s plan. Unilateral pacing by US labs would hand an advantage to rivals who keep accelerating, which is why the essay’s third point, global regulation, is doing most of the work and has the least mechanism behind it. Previous attempts at international AI coordination, including summit declarations signed in Paris and New Delhi earlier this year, produced commitments without enforcement.
The timing also raises the question of what a public Anthropic would owe shareholders if it deliberately slowed product releases to protect safety margins. A company valued near $2 trillion on projections of $200 billion in revenue by 2028 has investors expecting growth. Amodei’s essay asks them to accept a slower curve in exchange for reduced risk. Whether the market prices that trade is the test the IPO will set.