On 18 June 2026, the Electronic Frontier Foundation filed an amicus brief in Anthropic v. Department of War arguing the Trump administration’s export controls on Anthropic’s Mythos and Fable models are unconstitutional. The brief, authored by EFF attorneys Tori Noble and Corynne McSherry, makes three claims: the controls are retaliation for Anthropic’s refusal to let the Pentagon use its models for autonomous killing and mass surveillance; other AI models with similar capabilities face no equivalent restrictions; and the First Amendment precedent from the 1990s encryption wars — “code is speech” — applies directly to AI model exports.

This is the first major legal challenge to the Lutnick letter regime, and the first to frame the dispute as a constitutional rather than a regulatory question.

🔍 THE BOTTOM LINE

The EFF brief is significant for three reasons. (1) It is the first formal legal argument that the export controls are retaliatory — not security-motivated but punishment-motivated, which under First Amendment doctrine triggers strict scrutiny. (2) It establishes a direct precedent chain from the 1990s encryption export cases (Bernstein v. United States, Junger v. Daley) to the current AI export controls, which means the existing case law on software-as-speech is now in play. (3) It notes that other LLMs with similar offensive cybersecurity capabilities are not subject to export controls, only Anthropic’s, which undermines the government’s national-security justification. For New Zealand, the brief is the first document in the public record that frames the AI export-control dispute as a rights question rather than a trade question — and the rights framing is the one that small jurisdictions can use to argue for access.

What the EFF Argued

The brief, filed in Anthropic v. Department of War, lays out a timeline that the mainstream coverage of the Lutnick letter has not fully connected:

1. Anthropic refused the Pentagon’s demands. According to the EFF, the federal government relied on Anthropic’s models for use in classified systems for years. But after Anthropic resisted the government’s demands to use its models to “autonomously kill people or spy on Americans,” the government designated the company a “supply chain risk” — effectively banning agencies and government contractors from doing business with it. A court issued a preliminary injunction preventing those sanctions from taking effect, in a case where EFF and other civil liberties organizations filed an amicus brief earlier in 2026.

2. The export controls are the second wave. The Trump administration then imposed export controls that ban any foreign nationals from using Anthropic’s Mythos and Fable models. To comply, Anthropic shut the models down for everyone. The EFF argues these controls are a continuation of the same retaliatory pattern: the company refused to cooperate with the Pentagon’s autonomous-weapons programme, and the government responded by cutting off its revenue.

3. The retaliation is selective. Other LLMs with similar offensive cybersecurity capabilities are not subject to export controls. The government has instead embraced a voluntary system in which companies submit models to the government for cybersecurity testing 30 days before public release. Anthropic alone faces the mandatory regime. The EFF argues this selectivity is evidence of retaliatory motive, not security rationale.

4. “Code is speech” applies. The EFF’s most consequential argument is the precedent chain. In the 1990s, EFF challenged export controls on encryption software, helping establish the principle that “code is speech” protected by the First Amendment. Courts recognised that software is not just a functional tool — it is a means of expressing ideas, knowledge, and technical know-how. The EFF argues that AI models raise new questions but implicate the same constitutional and speech concerns as older efforts to restrict encryption. Export controls on AI models, like export controls on encryption, restrict the free flow of digital technologies and deprive the public of tools they need.

Why the First Amendment Framing Matters

The government’s defence of the Lutnick letter has been framed entirely in national-security terms. The Commerce Department letter warned that Anthropic’s models could be used to find and exploit software vulnerabilities, and that this capability constitutes a national-security risk if accessible to foreign nationals. The framing is the same one used for export controls on munitions, cryptographic hardware, and dual-use technology.

The EFF’s argument reframes the dispute. If the export controls are a content-based restriction on speech — software as expression — then the government must meet strict scrutiny, the highest standard in US constitutional law. Under strict scrutiny, the government must show the restriction is narrowly tailored to serve a compelling government interest. A selective restriction that applies to one company’s models but not to comparable models from other companies is, by definition, not narrowly tailored.

The brief also argues that even if the controls are not retaliatory, they are still unconstitutional because they are “unilaterally imposed without clear and fair standards.” The government has not published criteria for when an AI model triggers export controls, has not explained why Mythos qualifies but comparable models from OpenAI, Google, or xAI do not, and has not provided a process for challenging the designation.

The New Zealand Connection

The EFF brief is the first document in the US public record that frames the AI export-control dispute as a rights question. Every prior piece of coverage — the Lutnick letter, the Schneier/Sanders Guardian essay, the Macron G7 response, the NYT reporting on Anthropic employees — has framed the dispute as a trade, policy, or workplace question. The EFF brief frames it as a constitutional rights question.

For New Zealand, the rights framing is the one that small jurisdictions can use. Trade arguments require leverage at the WTO or in bilateral negotiations. Policy arguments require a seat at the G7 table. Rights arguments require a court. NZ’s legal system, like Australia’s and Canada’s, has a Bill of Rights Act that protects freedom of expression. If the US courts accept the EFF’s argument that AI model exports are speech, the precedent is exportable: a NZ court could find that restricting access to AI models on national-security grounds, without narrow tailoring, violates expression rights under the New Zealand Bill of Rights Act 1990, Section 14.

This is not a hypothetical. The Lutnick letter regime already restricts NZ researchers’ and companies’ access to Anthropic’s models. If the EFF wins the First Amendment argument in the US, the legal foundation for that restriction weakens in every Five Eyes jurisdiction that has an equivalent free-expression guarantee.

What to Watch

• Anthropic v. Department of War — the case is now live. The EFF amicus brief is filed. The government’s response will indicate whether it defends the controls as security-motivated (narrow scrutiny) or concedes the retaliatory framing (strict scrutiny).
• Other AI labs’ voluntary submissions — if OpenAI, Google, or xAI voluntarily submit their models to the government’s 30-day cybersecurity review programme, the selectivity of the Anthropic controls becomes harder for the government to defend.
• Open letter from cybersecurity experts — the EFF brief references a recent open letter from “leading cybersecurity experts and executives” arguing the sanctions prevent developers from using the best models to find and fix vulnerabilities. If that letter gains signatures, the security-motivated defence weakens further.

The EFF has been here before. The encryption wars of the 1990s ended with the government’s export controls on software declared unconstitutional. The AI export-control wars may end the same way. The question is how long it takes, and how many models are shut down in the meantime.