User:Louis/Anthropic and restrictions on who may use AI models

On June 12, 2026, Anthropic said it received a US government export-control directive at 5:21pm Eastern that ordered it to suspend all access to its Fable 5 and Mythos 5 models for any foreign national, whether inside or outside the United States, including the company's own foreign-national employees. Anthropic described the net effect of the order as forcing it to disable both models for every customer to stay in compliance. The directive landed two days after Anthropic chief executive Dario Amodei published an essay arguing that the US government should have the power to block or reverse the release of frontier AI models on public-safety grounds.

Anthropic's June 12, 2026 statement says a US government export-control directive, received at 5:21pm Eastern, suspended Fable 5 and Mythos 5 access for any foreign national inside or outside the United States.

Background

Anthropic's June 12, 2026 statement is the only primary account of the directive. In it, the company quoted the order's scope.

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.

Anthropic dated the page June 12, 2026 and said it received the directive that day at 5:21pm Eastern. June 12, 2026 was a Friday. The company said the order left it no compliant option other than disabling Fable 5 and Mythos 5 for all customers. News coverage followed on June 13, 2026.

The government gave no specific account of its concern. Anthropic wrote that the letter did not provide specific details of its national security concern, and that the company's understanding was that the government believed it had become aware of a method of bypassing, or jailbreaking, Fable 5. No public copy of the letter exists. There is no Federal Register entry, no Bureau of Industry and Security notice, and no PDF; every fact about the order traces to Anthropic's characterization or to journalists' reporting.

The sender is named only in degrees. Anthropic's own statement attributes the directive to the US government and names no agency. CNN and Fortune reported that the Commerce Department issued the restriction. NBC News reported that the letter was sent from Commerce Secretary Howard Lutnick and was written with the help of officials from the Commerce Department's Bureau of Industry and Security, citing an administration official; those identifications appear only in reporting, not in any primary document. Anthropic framed the order as an export-control directive citing national security authorities, and named no statute.

This directive is separate from a longer-running dispute between Anthropic and the Department of War over military use of Anthropic's models.

The administration's account

David Sacks, the White House AI and Crypto Czar, gave the administration's version of the sequence in a June 13, 2026 post on X, the day after Anthropic disabled the models. The post is Sacks's personal statement of what he believes to be true, not an official government record, and the claims in it are contested.

Sacks wrote that Fable is Mythos with guardrails, and that if those guardrails fail the result is to expose Mythos and its cyber capabilities to people who should not have them. He wrote that a trusted partner of both Anthropic and the US government, while testing Fable, came forward with a jailbreak of those guardrails; that the administration asked Amodei to fix the jailbreak or de-deploy the model; and that Amodei refused. By Sacks's account the administration then issued the export control in reaction, did so reluctantly, and hoped that Anthropic would remediate the issue so the control could be lifted and Fable returned to general release.

Sacks disputed Anthropic's characterization of the bypass. He wrote that Anthropic's blog post defended its decision by saying the jailbreak isn't serious, and that this was not what the trusted partner and the government believed. He wrote that attempts to tie the export control to the earlier Department of War dispute are wrong, and that the administration values Anthropic's technical capabilities and regards the issue as one that should be easily resolved.

Anthropic's published statement gives a different account of the same bypass. In its June 12, 2026 statement, Anthropic said it had reviewed a demonstration of the technique and identified a small number of previously known, minor vulnerabilities, that other publicly available models could discover the same flaws without a bypass, and that it disagreed that the finding of a narrow potential jailbreak should be cause for recalling a model deployed to hundreds of millions of people. Anthropic said it was complying with the directive while calling the matter a misunderstanding it was working to resolve.

Dario Amodei's position on government control of AI models

Dario Amodei's June 10, 2026 essay argues that frontier AI models should undergo mandatory testing and that the government should be able to block or reverse their release on public-safety grounds.

Two days before the directive, on June 10, 2026, Amodei published an essay titled Policy on the AI Exponential. In it he argued for binding government authority over frontier model releases.

Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety.

He also wrote that the government should have the power to block or deter deployment of a model if it is determined, based on third-party assessment, to present unacceptable risks. Axios reported the essay the same day, summarizing that the government should legally be able to block or deter dangerous AI deployments.

Amodei has a documented record backing access controls. In Machines of Loving Grace (October 2024) he described an entente strategy of blocking or delaying adversaries' access to key resources like chips and semiconductor equipment. In On DeepSeek and Export Controls (January 2025) he wrote that export controls are essential to keeping democratic nations ahead in AI development. In a January 6, 2025 Wall Street Journal op-ed with Matt Pottinger, he wrote that export controls have been a valuable tool in slowing China's AI development.

Amodei frames the June 2026 essay as a change from Anthropic's earlier posture. Anthropic's Responsible Scaling Policy describes itself as risk governance that should be proportional, iterative, and exportable. In the June 2026 essay, Amodei argued that transparency requirements alone are no longer sufficient.

Amodei asked for government authority to block or reverse a model's release, and within two days a government directive ordered Anthropic to suspend access to two of its own models. The policy came back to bite its author. The man who wanted the state to hold a veto over model releases watched it reach for one against his own product.

The dispute with the Department of War

Congressional Research Service product IN12669 describes the months-long dispute between the Department of Defense and Anthropic and the July 2025 contracts of up to $200 million each.

The directive is not Anthropic's only friction with the federal government. The company is in a public dispute with the Department of War (the secondary title adopted September 5, 2025 by Executive Order 14347 for the Department of Defense, which remains the department's statutory name) over use restrictions on Anthropic's models. The Congressional Research Service product IN12669 describes the matter as a reportedly months-long dispute between the Department of Defense and Anthropic over Department use of Anthropic products, including the company's generative AI model, Claude.

Executive Order 14347 authorizes Department of War only as a secondary title and states that statutory references to the Department of Defense remain controlling until changed by law.

Executive Order 14347 authorizes the Secretary of Defense the use of the additional secondary title, the Secretary of War, in official correspondence, public communications, ceremonial contexts, and non-statutory documents within the executive branch. The order states that statutory references to the Department of Defense shall remain controlling until changed subsequently by the law. Military.com reported in October 2025 that only Congress can amend or replace statutory titles.

Anthropic refuses two use cases. In a February 26, 2026 statement, Amodei identified mass domestic surveillance and fully autonomous weapons as the two cases the company declined to provide.

Two such use cases have never been included in our contracts with the Department of War, and we believe they should not be included now.

In a February 26, 2026 statement, Anthropic identified mass domestic surveillance and fully autonomous weapons as the two use cases it declined to provide to the Department of War.

Amodei gave the company's reasons: that frontier AI systems are not reliable enough to power fully autonomous weapons, and that using these systems for mass domestic surveillance is incompatible with democratic values. He wrote that the company cannot in good conscience accede to the request. As Anthropic quotes the demand, the Department of War stated it will only contract with AI companies who accede to any lawful use and remove safeguards in the cases mentioned above. The Congressional Research Service frames the same demand as all lawful purposes.

The timeline runs from contracts to threats to a refusal. In July 2025, the Department awarded contracts to Anthropic, Google, OpenAI, and xAI for up to $200 million each. On February 24, 2026, CBS News and the Washington Post reported that Secretary Pete Hegseth made a formal demand for full access for all lawful purposes. NPR's The Public's Radio reported a threat to terminate the $200 million contract, and the Washington Post reported a threat to invoke the Defense Production Act. Amodei issued the public refusal on February 26, 2026. On February 27, 2026, the Department designated Anthropic a supply chain risk, which Anthropic describes as a label reserved for US adversaries, never before applied to an American company.

That same day, the Congressional Research Service reports, President Donald Trump directed federal agencies to immediately cease all use of Anthropic's technology.

The dispute is in litigation on two tracks. At the district level, the Civil Rights Litigation Clearinghouse records Anthropic PBC v. U.S. Department of War et al., No. 3:26-cv-01996 (N.D. Cal.), before Judge Rita F. Lin, filed March 9, 2026, with a preliminary injunction granted March 26, 2026; the reported counts include Administrative Procedure Act violations, First Amendment retaliation, ultra vires action, and Fifth Amendment due process. The Clearinghouse summary states that in granting the injunction the court found Defendants had as of yet failed to prove that Anthropic's conduct qualifies as a supply-chain risk. On appeal, the matter is before the Ninth Circuit, No. 26-02011. The Congressional Research Service reports that the court of appeals denied Anthropic's motion for a stay on April 8, 2026, undoing the district court's injunction.

The Civil Rights Litigation Clearinghouse summary of Anthropic PBC v. Department of War records the March 9, 2026 complaint, its counts including First Amendment retaliation, and the preliminary injunction granted March 26, 2026.

What a nationality gate would require

Enforcing a rule of no foreign nationals requires checking identity and proof of citizenship at the point of access. The directive's scope reaches any foreign national, whether inside or outside the United States, so the only way to keep a foreign national out of a product is to make every user prove who they are first. That is the same show-your-ID mechanism as the age-verification laws I oppose. The Electronic Frontier Foundation's December 2025 review reports that half of the United States now mandates age verification for accessing adult content or social media platforms, and that nine states saw their laws take effect in 2025 alone. The National Law Review reported in January 2026 that roughly half of US states mandate some form of age gating, and that eight states have enacted laws that either ban minors from obtaining social media accounts outright or require parental consent. The Supreme Court upheld Texas House Bill 1181 in Free Speech Coalition, Inc. v. Paxton, No. 23-1122, decided June 27, 2025, holding that the law triggers, and survives, review under intermediate scrutiny because it only incidentally burdens the protected speech of adults. I oppose mandatory age-verification and show-your-ID mandates on privacy grounds, a position I have documented on my channel.

The Electronic Frontier Foundation's 2025 review reports that half of US states now mandate age verification for adult content or social media and that nine states' laws took effect in 2025.

The Supreme Court syllabus in Free Speech Coalition v. Paxton holds that Texas House Bill 1181 triggers and survives intermediate scrutiny because it only incidentally burdens adults' protected speech.

A nationality gate is defeated the same way a VPN ban is defeated. Access to these models is account-based, so a US account holder can hand credentials to someone overseas. The gate then burdens every ordinary user with an identity check while doing little to stop the targeted foreign use it was built to block.

A nationality gate would bar many of the people who built the models. The White House Council of Economic Advisers AI Talent Report, dated January 14, 2025, states that among PhD graduates the share of non-US citizens has exceeded 50 percent since 2003 and reached 59 percent in 2022. The MacroPolo Global AI Talent Tracker, which defines top-tier researchers by NeurIPS authorship, reports that US institutions employ 59 percent of the world's top-tier AI researchers. A gate that excludes foreign nationals would shut out a large share of the foreign-born researchers who built the field.

The White House Council of Economic Advisers AI Talent Report states that the non-citizen share of AI-relevant PhD recipients stayed above 50 percent from 2003 and reached 59 percent in 2022.

The MacroPolo Global AI Talent Tracker reports that US institutions employ 59 percent of the world's top-tier AI researchers and that this lead rests almost entirely on foreign-born talent.

See also

• Anthropic
• Age verification
• Right to repair
• DMCA Section 1201
• User:Louis/Alliance for Automotive Innovation and Section 1201