Anthropic's Amodei Calls for AI Slowdown, Days Before $2 Trillion IPO

Dario Amodei is asking the industry he helps lead to slow down, just as his own company prepares to go public in what could be the largest IPO in history. The Anthropic CEO published an essay on Saturday, September 13, titled "We Must Pace the Frontier," laying out a three-part plan to rein in the pace of frontier AI development.

The first commitment is internal: Anthropic will give third-party evaluators "permanent, employee-level access" to its systems, letting outside reviewers verify safety measures, flag incidents, and monitor alignment throughout training. The second part asks every frontier AI company based in a democratic country to adopt shared safety standards and caps on how fast unchecked capability gains can proceed.

The third and most ambitious piece calls for coordination between democratic AI powers and autocracies, China chief among them, potentially through a narrow agreement barring the most obviously dangerous applications, such as using AI to help produce biological weapons. Anthropic said it has already caught users attempting to route around its safeguards for exactly that purpose.

Amodei's essay includes a stark warning: within six to twelve months, he argues, coordinated swarms of AI agents could become capable of "taking over the entire internet." Sam Altman responded within hours, posting that he agreed AI development "needs to pace the frontier" and pledging that OpenAI would grant its own independent evaluators similar employee-level access.

Altman also confirmed OpenAI will not pursue a public listing in 2026, attributing the decision to unresolved safety concerns.

Not everyone is convinced

The reaction has been far from uniform. Some AI safety researchers say the plan doesn't go nearly far enough, and several specifically pushed back on the internet-takeover scenario, calling it "not particularly plausible."

Jacob Coxon, a researcher who recently resigned from Anthropic, offered a sharper indictment, accusing both Anthropic and OpenAI of "racing straight to self-improving superintelligence and gambling with our lives" even as their leaders talk publicly about caution.

"The people building AI earnestly believe that it could kill us all by the end of the decade," Coxon warned.

President Trump dismissed the warnings altogether, describing them as the work of "very negative forces" pushing "things that won't happen." He reiterated that "we're leading China in AI" and that "whoever wins AI, wins," a stance that sits awkwardly next to Amodei's call for slower, more coordinated progress.

The timing complicates how the essay is being read. Anthropic is reportedly gearing up to market a stock offering valued above $2 trillion, a listing that would dwarf any IPO to date, raising questions about whether a company preparing for that scale of investor enthusiasm can credibly champion restraint at the same moment.