Lutnick orders Anthropic to cut off Fable and Mythos

Anthropic said Commerce Secretary Howard Lutnick's directive reached the company at 5:21 p.m. ET on June 12, 2026. The letter ordered suspension of Fable 5 and Mythos 5 access for any foreign national, whether that person lived abroad or worked inside the United States.

Anthropic disabled the models for all customers in all regions because its account systems couldn't verify nationality in real time. Amazon Web Services revoked access across its regions after Anthropic asked the cloud provider to implement the cutoff.

Anthropic launched Fable 5 on June 9, 2026 as a Mythos-class model with broader developer access. Mythos 5 stayed restricted through Project Glasswing, a program limited to cyber defenders, infrastructure providers, and selected biology researchers. Anthropic priced Fable 5 at $10 and $50 per million input and output tokens, putting it in the high-end commercial tier rather than consumer pricing.

Administration officials told Anthropic that a possible jailbreak of Fable 5 had surfaced a small number of known, minor software vulnerabilities. Anthropic said the demonstration found no universal bypass and that other publicly available models could find the same vulnerabilities without a bypass. The company said it had run thousands of hours of internal safety testing before launch.

Business Insider reported that White House and Cabinet officials spent about 24 hours pressing Anthropic to pull the models voluntarily before Commerce imposed the directive. The reported calls involved Amazon CEO Andy Jassy, Treasury Secretary Scott Bessent, White House cyber director Sean Cairncross, White House chief of staff Susie Wiles, Commerce's Howard Lutnick, and Commerce undersecretary Jeffrey Kessler, alongside Anthropic CEO Dario Amodei.

Anthropic-side sources told Business Insider the company received about 90 minutes to comply after officials declined to share detailed technical evidence. That sequence puts the order in a process dispute, not only a cybersecurity dispute, because officials converted a contested risk assessment into an immediate nationality-based cutoff.

Congress passed the Export Administration Act of 1979 to replace the lapsed 1949 Export Control Act and to put Cold War export rules under a renewable statute. The current Export Administration Regulations in 15 C.F.R. Parts 730-774 now sit under International Emergency Economic Powers Act authority, because the 1979 statute lapsed in 2001 and Congress has not renewed it. The Bureau of Industry and Security has run the program continuously under that emergency authority for more than two decades.

15 C.F.R. Part 734 treats certain in-country releases of controlled technology or source code to a foreign person as exports to that person's country of citizenship. BIS explains that in-country release to a foreign person counts as a deemed export. The hard question for the Fable and Mythos order is whether a live API call to a frontier model is the kind of controlled technology the 1979 Congress meant to cover.

BIS has used deemed-export rules for decades in university and lab settings, most visibly in cases involving graduate students and visiting researchers with controlled semiconductor or cryptography know-how. The Anthropic cutoff is the first use BIS has acknowledged in a live commercial AI service. Reuters noted that U.S. AI controls have mostly hit chips and tools that support AI training, not access to a deployed model. The directive moved the pressure point from hardware to a software-service endpoint where the user, the employee's nationality, and the data center can sit in three different jurisdictions.

The foreign-national rule inside the United States caught H-1B workers, F-1 STEM OPT students, and visiting researchers even when they worked for American companies. USCIS data showed roughly 740,000 H-1B workers and more than 200,000 STEM OPT participants in active status as of late 2025. Existing BIS guidance exempts U.S. citizens, lawful permanent residents, and protected individuals, but ordinary API access doesn't run a citizenship check at every prompt.

That compliance gap explains why Anthropic chose to revoke access for all customers rather than build a nationality screen for every query. The same gap means a U.S. employer with foreign-national engineers can't promise those engineers continuous access to frontier AI tools, even inside a U.S. office.

European Commission spokesperson Thomas Regnier told reporters the EU was examining the practical consequences of Anthropic's cutoff and warned that AI safeguards shouldn't discriminate against close partners. Canadian Prime Minister Mark Carney used the G7 summit in Alberta to argue that allied countries shouldn't rely on a single AI provider and noted that more than 70% of Canadian exports still go to the United States.

Anthropic's launch materials described internal protein-design work that ran about 10 times faster with Mythos 5 than with prior models. That biological-design capability is part of why a model restricted to Project Glasswing drew the kind of export-control response usually reserved for dual-use laboratory equipment.

The public record on June 14, 2026 confirmed the directive through Anthropic, Reuters, and administration-source reporting, but didn't include a published Federal Register rule or public BIS classification for ordinary API access. Congress can demand the directive, the technical basis, and the agency's legal theory. Courts can review the order only if Anthropic or an affected user brings a challenge with standing and a reviewable agency action.