The export-control order landed on a Friday in June with little warning. U.S. officials told Anthropic to cut off foreign access to its two newest models, Fable 5 and Mythos 5, citing national security concerns. Three weeks later, the government is preparing to walk that back.
The restoration of Fable 5, the flagship model at the center of the dispute, is expected within days, according to people familiar with the matter. The precise timing and any conditions attached were still being settled this week, the people said. Anthropic declined to comment on the pending decision.
The June 12 directive was delivered as an emergency instruction, according to a person close to the company. It required Anthropic to block access for foreign users immediately, including customers reached through its cloud partners, and gave the company little time to warn clients before the switch was flipped. Some customers learned their access had been cut when their applications began returning errors.
The timing was awkward for Anthropic. The company closed a $65 billion H-round in late May at a valuation of roughly $965 billion, making it the most valuable AI startup in the world. Enterprises across Europe, the Middle East and Asia had built workflows around Fable 5, and the abrupt shutdown forced some to scramble for alternatives.
The episode rattled markets. Shares of AI and computing names wobbled in the days after the order as investors weighed what a permanent restriction would mean for the industry’s central revenue engine: selling model access to customers around the world. Some foreign customers, according to people familiar with the matter, had already begun testing rival systems from OpenAI and Google as insurance.
Government officials have offered few public details about why the restriction is being lifted. A person close to the company said the talks settled on a review process that would allow foreign customers to regain access after vetting, rather than a blanket ban. The arrangement, if it holds, would formalize a practice that until now has been ad hoc.
The case fits a broader pattern. Washington has spent the past year deciding how far to extend export controls — long applied to chips — to the models themselves. Frontier labs now brief the government before major releases. OpenAI, for instance, consulted U.S. officials ahead of the June 27 preview of its GPT-5.6 line and agreed to limit early access to a small group of vetted partners.
For Anthropic, the stakes go beyond a single product. The company’s valuation rests on the assumption that frontier models can be sold globally at scale. Analysts said a pattern of national-security-driven shutdowns would force customers to price in political risk, slowing adoption of high-end models in allied markets the U.S. wants to influence.
Industry executives say the uncertainty cuts both ways. A permanent ban would have pushed the most capable American models out of allied markets and toward Chinese competitors offering their own frontier systems. The pending restoration reflects a judgment that controlled access beats no access, they argue.
What remains unclear is whether Mythos 5, the second model covered by the June order, returns on the same terms. People familiar with the matter said the two models were being handled separately, and that Mythos 5’s fate could take longer to resolve because of its use in high-end research settings.
The episode has left a residue of caution among customers. Procurement officers at two European technology firms, according to people familiar with the matter, are now asking model vendors for contractual guarantees about continuity of access — a clause that barely existed a year ago.
Anthropic executives are trying to frame the episode as evidence of maturity: a young company navigating hard national security questions alongside a government that takes them seriously. Whether investors accept that framing will show in the next funding discussion, analysts said.
The broader question is whether Washington settles on a durable framework for who may use the most advanced models, and where. The Fable 5 case suggests the answer is being written case by case, with each new model launch becoming a negotiation in itself.
For now, the message to foreign customers is that access can be granted and taken away at the stroke of an order. That is a fragile basis for a global business, analysts said, and Anthropic’s answer — to negotiate its way through each episode — may be the only one available while policy is still finding its footing.
The next test will come with the models already in development. People familiar with the company’s roadmap say Anthropic plans to ship two more frontier systems before the end of the year, each of which will face the same review. Whether those launches run smoothly may determine whether the current episode is remembered as a disruption or as the start of a workable routine.
