By Christian Martin
Anthropic wants the world’s top labs to consider slowing down. The plea arrives with a confidential IPO filing under its arm and a near-trillion-dollar valuation on its books. The timing is not incidental.
Anthropic, now officially the most valuable startup on the planet with a near-$1 trillion price tag and a confidential IPO filing tucked under its arm, has just delivered a dramatic plea: the world’s top AI labs should consider slowing down, maybe even pausing altogether, because models are getting dangerously close to "recursive self-improvement." The company’s new blog post, co-authored by head of research Marina Favaro and co-founder Jack Clark, warns that AI systems capable of upgrading themselves without human intervention could arrive "sooner than most institutions are prepared for" and that a global slowdown would "likely be a good thing."
It is a striking message. It is also exquisitely timed.
Not long ago, the same company published a very different kind of paper, one you will not see quoted in this week’s safety sermon. In its 2028 AI Leadership document, still live on its own site, Anthropic advocated unequivocally for United States global dominance in AI, framing the race in almost martial terms and urging policies to cement American primacy. The tension between that call and this week’s plea for a global pause could hardly be starker: one hand is waving a flag to sprint ahead, the other is urging everyone to freeze in place. Which is it: leadership by acceleration, or safety by deceleration? Or is the real message simply, "slow down after we have won"?
That question hangs over every paragraph of the company’s latest PR push. As the Wall Street Journal reported, Anthropic’s run rate is on track to hit $50 billion in annualized revenue by the end of this month, up from $9 billion at the close of 2025. The money is real, the IPO is looming, and suddenly the firm that built its brand on existential caution wants the whole world to gather around the table and talk about pumping the brakes. If you are a developer or enterprise buyer already grappling with Anthropic’s steadily climbing API prices and ever-tighter rate limits, the irony is inescapable: the company wants a global slowdown while it monetises access as aggressively as any hyperscaler.
David Sacks, venture capitalist and Trump adviser, has long dismissed Anthropic’s safety advocacy as "regulatory capture." This week’s performance will not change his mind. Skeptics can point to the decision to limited-release a powerful "Mythos" cybersecurity model (capable of finding bugs and vulnerabilities) as a classic marketing teaser. You do not have to be a cynic to see the pattern: publish scary warnings, dangle a restricted product that only fuels the mystique, and then stand before Washington and the public as the responsible adult in the room.
Anthropic built its brand on existential caution, even as its valuation and revenue accelerate.
The company’s own numbers tell a story of breakneck acceleration, not restraint. A $1 trillion valuation, $50 billion annualized revenue, a confidential IPO filing: these are not the hallmarks of a lab preparing to voluntarily slow down. They are the hallmarks of a company racing to lock in a market position before the public markets weigh in. And that is where the pivot to a "global agreement" on AI speed becomes so conveniently useful. If you can persuade policymakers that the frontier is too dangerous to approach without guardrails, you simultaneously set the terms of engagement, and potentially freeze competitors who do not have a $1 trillion valuation to cushion the pause.
Anthropic’s post itself acknowledges the deadly flaw in any slowdown fantasy: "Training runs are far easier to conceal than missile silos," and "whoever continues while others pause could inherit the lead." Coming from a firm that recently championed US leadership, this reads less like a sincere proposal for collective restraint and more like a warning to rivals: we are ready to dominate if you stop. The idea of a verified global slowdown with teeth, while the US simultaneously pursues AI supremacy, collapses under its own contradictions. Either you want a competitive moat, or you want a global safety architecture. You cannot have both while telling developers they have hit their rate limit.
To be fair, Anthropic has plenty of genuine safety researchers. Wharton’s Ethan Mollick describes the company as a "mix of things": a trillion-dollar marketing machine, a pack of relentless model builders, and a cadre of "philosopher kings" who truly fear what they are building. But when a company speaks publicly, it is the trillion-dollar machinery that chooses the timing and the message. And right now, with an IPO on the horizon and revenue exploding, a call for a global pause is the ultimate credibility play: it says we are so powerful that even we are scared. That is a brilliant story to sell to investors, and it also happens to kneecap anyone who cannot afford to slow down.
Anthropic’s sudden desire for a world summit on AI velocity does not nullify their earlier call for US dominance; it reframes it. The dominance paper was the strategy. The pause paper is the shadow roadshow. Developers feeling the pinch of rising costs and tighter access should pay close attention: the company that is asking the world to slow down is the same one that is pricing you out of the fast lane.
Sources: Anthropic, "2028 AI Leadership"; and The Wall Street Journal, "Anthropic Urges Global Pause in AI Development, Flags Self-Improvement Risk".