Judge Rules Pentagon's Blacklisting of Anthropic Was First Amendment Retaliation

What happened

Judge Rita F. Lin of the Northern District of California ruled that the Pentagon's blacklisting of Anthropic was unlawful First Amendment retaliation. She vacated the supply-chain-risk designation and the ban on federal agencies and defense contractors using the company's products, ordering the administration to rescind its directives.

Context

Anthropic filed suit in March 2026 after Defense Secretary Pete Hegseth designated the company a supply-chain risk, a step he took following Anthropic's refusal to lift its restrictions on mass surveillance of Americans and lethal autonomous weapons. Trump and Hegseth then ordered all federal agencies to permanently stop using Anthropic's products and banned every defense contractor from transacting with the company, even on non-military work. Hegseth simultaneously renegotiated all AI lab contracts with the military to permit use for any lawful purpose and signed replacement deals with seven other labs, including Google, Microsoft, OpenAI, and SpaceX.

How it works

The supply-chain-risk designation is a Department of Defense mechanism that, once applied, bars a vendor from federal contracts and prohibits defense contractors from transacting with them. It requires no congressional vote and no adversarial hearing. The government's legal case rested on a claim that Anthropic retained backdoor access to its technology once deployed inside a national security system. Judge Lin found that claim unsupported. She found the administration had provided only a thin justification before retreating from the core of its own risk assessment, and she ruled that Anthropic's closed models pose no greater national-security risk than any other opaque AI system. The designation, in her view, was arbitrary and capricious, and the First Amendment barred its use as a penalty for Anthropic's public criticism of the administration.

Our read

The most consequential line is not the First Amendment holding. It is the finding, from the government's own records, that the designation was triggered by Anthropic's public communications. A statement of motive, in writing, in the agency's file. That makes the national-security framing a wrapper around a speech-based penalty.

The renegotiation of every AI lab's contract to permit any lawful use reveals the actual goal: strip use restrictions from the entire vendor base. Anthropic refused, so Anthropic got blacklisted. The seven replacement deals with Google, Microsoft, OpenAI, and SpaceX are not a competitive procurement. SpaceX is not primarily an AI lab.

The court's finding that opaque AI is the baseline risk, not a unique threat, also undercuts the legal theory for applying supply-chain designations to AI vendors on model opacity alone. If a black box is the default, the designation needs a more specific justification.

For a studio running ComfyUI with local models, this validates the default: local open-weight inference creates no supply-chain compliance exposure. What changes is the removal of the federal ban on Claude API, relevant if that feeds a prompt pipeline.

What this changes

If the studio's pipeline includes Claude API for script drafting or prompt engineering, the federal prohibition no longer applies. No configuration change is needed on the local ComfyUI stack; the ruling does not touch open-weight model licensing, inference hardware, or the ComfyUI pipeline. For a studio with no Anthropic dependency, nothing changes on Monday. The practical impact concentrates on defense contractors and federal agencies, where the rescinded directives make Anthropic eligible for procurement again. Whether the Pentagon re-engages is a separate question, and the replacement contracts with the other labs remain in force absent an appeal.

License

No software licence applies. This is a court ruling, not a model or tool release. The sources do not state a licence for Anthropic's weights or the Claude API, and that question is unchanged by this decision.

Key takeaways

• The court found the supply-chain-risk designation was retaliation for Anthropic's public stance, citing the agency's own records as evidence of motive.
• The government's backdoor-access claim was found unsupported; the court ruled Anthropic's models are no greater a risk than any other opaque AI system.
• The renegotiation of all AI lab contracts to permit any lawful use shows the goal was removing restrictions broadly, not addressing a specific Anthropic vulnerability.
• For local-model studios running ComfyUI, nothing changes operationally; the ruling lifts the federal ban on Claude API but does not affect open-weight model usage.
• The replacement contracts with seven other AI labs remain in force; the ruling vacates the Anthropic-specific ban, not the broader procurement shift.

Sources

1. Anthropic was illegally blacklisted by the Trump administration, court rules — tier 2
2. Trump blacklisting of "woke" Anthropic deemed illegal by federal judge — tier 2