Anthropic calls for global AI freeze while mending White House ties ahead of IPO

Anthropic's global pause call

Anthropic, the San Francisco-based AI company valued at roughly $1 trillion, on Thursday proposed a globally coordinated slowdown or temporary halt to frontier AI development. The company argued in a blog post and a report from its Anthropic Institute that such a pause would "enable societal structures and alignment research to keep up with the advance of the technology." The call echoes a 2023 open letter signed by Elon Musk, but Anthropic insists any pause must be multilateral, verifiable, and with clear triggers—otherwise it would simply hand leadership to whoever keeps going.

We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development.

The self-improvement risk

The specific trigger Anthropic identifies is recursive self-improvement—AI systems that can autonomously design and build their own successors. The company warns that this milestone "could come sooner than most institutions are prepared for" and that "full recursive self-improvement might increase the risks of humans losing control over AI systems." As a marker of how close the loop already is, Anthropic disclosed that as of May, over 80% of the code merged into its own codebase was written by its model Claude. "If a model already writes the overwhelming majority of the code that builds the next model, the loop between a system and its own improvement is no longer theoretical, merely partial," the company said.

A coordination puzzle

A key problem is enforcement. Without a global mechanism that includes rules, verification, and an oversight body, any pause would be toothless. Anthropic acknowledges that none of this infrastructure exists, and that its direct competitors—many of whom are pouring hundreds of billions into AI datacentres—have shown no appetite to slow down. The company argues that only an agreement among multiple leading labs in multiple countries, "most notably the United States and China," could work.

Without a global coordination mechanism, companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures.

The Anthropic Institute says it will now work with collaborators to design the verification systems a credible pause would require.

Thaw with the White House

The pause proposal arrives as Anthropic works to repair a fractious relationship with the Trump administration ahead of its anticipated public listing. In March, the Defense Department labeled the company a "supply-chain risk" after it refused to allow its AI to be used for domestic surveillance and fully autonomous weapons. But tensions have eased since CEO Dario Amodei visited the White House in mid-April. The administration even invited Amodei to a May 21 signing of an AI executive order, though Trump canceled the event; he signed the order on Tuesday, and Anthropic said it looked forward to "collaborating" on its implementation. The company has also discussed with National Cyber Director Sean Cairncross ways its powerful Mythos model could protect critical infrastructure. Still, the supply-chain designation remains in place, and the Pentagon is "vigorously" defending itself against Anthropic's lawsuit.

Anytime the government signals that it's washing its hands of a company, that's a major problem for that company.

1. 2026-03Defense Department designates Anthropic a 'supply-chain risk', barring military contractors from using its AI.
2. 2026-04CEO Dario Amodei visits White House to discuss collaboration, first meeting since dispute erupted.
3. May 21, 2026White House planned AI executive order signing, with Amodei invited, but canceled due to Trump's objections.
4. Jun 2, 2026Trump signs revised AI executive order; Anthropic says it looks forward to collaborating.
5. Jun 5, 2026Anthropic proposes a global, verifiable pause in frontier AI development.

Pushback and skepticism

The proposal has already drawn fire from both industry rivals and some White House officials, who argue that using safety to slow down is a cover for hobbling competition under the guise of protecting the public. Critics also point to Anthropic's selective release of Mythos—which is currently limited to a small number of vetted organisations due to its cybersecurity capabilities—as a marketing tactic to hype its products. Anthropic, however, maintains that its warnings are grounded in research from the newly created Anthropic Institute, whose mission is to "tell the world" what it learns about risks as AI advances. Whether a pause can be agreed upon before the self-improvement feedback loop tightens further remains entirely open.