Anthropic Blacklisting Blocked, For Now: What The Anthropic Injunction Means — And What It Doesn't — For AI Businesses

Update: After the publication of this article, the government appealed the District Court’s preliminary injunction order to the U.S. Court of Appeals for the Ninth Circuit (Case Number 26-2011). And, on April 6, 2026, the government filed a Status Report (Dkt. No. 146) in accordance with the District Court’s Preliminary Injunction Order describing the steps it has taken to ensure compliance with the preliminary injunction. The Status Report outlines how the government has circulated the preliminary injunction to all agencies, formally instructed personnel to stop implementing the enjoined directives, rescinded prior guidance restricting Anthropic, restored access to and use of Anthropic products across systems and contracts, and directed contractors to reinstate Anthropic services. It also represents that agencies have implemented these measures and certifies compliance with the injunction.

On March 26 in the Northern District of California, the federal district court overseeing one of the two Anthropic versus the U.S. Department of War (DoW) cases granted a preliminary injunction blocking the enforcement of a set of what the court called “Orwellian” government actions that would have effectively blacklisted the AI company Anthropic from federal contracting by labeling it a national security “supply chain risk.”¹ 

Of particular note, the court rejected an argument by the DoW that “any vendor who ‘pushes back’ on or ‘questions’ DoW becomes an adversary” as “deeply troubling and inconsistent with the statutory text.”² The court made clear that the governing law “does not permit DoW to designate IT vendors as ‘supply chain risks’ whenever they ask probing questions or stubbornly insist on particularly contracting terms, even if their doing so causes DoW to doubt their trustworthiness.”³ A parallel case involving related government actions is still pending in the D.C. Circuit Court.⁴

The ruling represents an important early victory for Anthropic and for AI companies generally to the extent it indicates they have protection from retaliatory agency actions when they engage publicly on safety and policy issues related to the government’s use of their products and services. That said, the order expressly does not require the DoW to buy Anthropic’s products or to accept Anthropic’s preferred contract limitations. It also remains the case that AI businesses and businesses that use AI need to carefully manage their relationships and interactions with the U.S. government and take the protective measures we discussed in our article last week: HSF Kramer Insights, ‘All lawful uses’: Precautions AI businesses need to take after Anthropic v. US DoW.

What the court actually did

The court’s preliminary injunction in favor of Anthropic is an interim remedy that temporarily pauses the government’s designation of Anthropic as a supply chain risk until the court can hear evidence and issue a final ruling. 

In deciding to issue the injunction, the court found that Anthropic was likely to succeed on each of its three theories:

• First Amendment retaliation. The court cited evidence that the challenged actions were motivated by hostility toward Anthropic’s public statements about AI safety and military use, rather than by legitimate supply chain concerns. In particular, the court homed in on DoW Secretary Pete Hegseth’s public statements that “expressly tied Anthropic’s punishment to its attitude and rhetoric in the press.”⁵
   
• Procedural due process violations. The court agreed that the government imposed sweeping, reputationally damaging exclusions without notice or an opportunity to respond.⁶ It called out that the government’s actions not only threatened to “cripple” Anthropic by “strip[ping]” it “of billions of dollars in federal contracts and subcontracts,” but also through reputational harm “by labeling it as an adversary to the United States and ending its ability to have any commercial relationship with any company that might want to do business with DoW.”⁷ It also emphasized the fact that Anthropic is a U.S.-based company, distinguishing it from foreign companies that do not rely as heavily on federal contracting, where less notice might be warranted.⁸
   
• Administrative Procedure Act violations. The court agreed that the DoW likely exceeded its statutory authority and did not follow required procedures, including by failing to consider less intrusive measures and justifying its decision with reasons that ran counter to the evidence.⁹ The court was especially critical of the government’s apparent position “that any vendor who ‘push[es] back’ on or ‘questions’ DoW becomes its ‘adversary,’” labeling it “deeply troubling” and “inconsistent with statutory text.”¹⁰

The court also agreed that the timing of the government’s decision was suspicious, given that the evidence showed “in the parties’ multi-year working relationship and during Anthropic’s extensive national-security vetting process,” the government “never raised any supply chain concerns[.]”¹¹ It highlighted that nothing in the record supported the government’s position that Anthropic would have the technological ability to cause its software to stop working once deployed on national security systems.¹² And it rejected the argument that Anthropic could introduce hard-to-detect vulnerabilities, in light of the “opaque” nature of its AI technology, as “inherent to AI technology.”¹³ 

What the ruling does not do

• The court did not reach a final determination on the merits of the case. Instead, it found only that Anthropic’s challenges were likely to succeed on the merits.
   
• The court did not establish a general rule limiting the government’s ability to impose supply chain risk designations or other exceptional procurement restrictions in materially different factual situations, such as those involving non-U.S. companies or companies that host government data in systems under their control.
   
• The court expressly did not compel the U.S. government to contract with Anthropic. The court was clear that it remains the DoW’s “prerogative to decide what AI product it uses.”¹⁴
   
• The court’s decision does not prevent agencies from making ordinary contracting decisions, setting technical requirements or declining to contract with companies whose products do not meet their needs.

What happens next

• The DoW has requested and received a seven-day administrative stay to allow it time to seek an emergency stay from an appellate court, which as of publication of this article had not yet been sought.
   
• The DoW’s supply chain designation was also made pursuant to a second statute, and Anthropic challenged that designation in a federal appellate court in Washington, D.C. See Anthropic PBC v. United States Department of War, No. 26‑1049 (D.C. Cir.). In order for Anthropic to avoid the impending business impact of the “supply chain risk” designation, it needs both the California decision to stand and the Washington D.C. court to rule in its favor. Following the California court’s preliminary injunction order, Anthropic filed a letter in the D.C. court arguing that the ruling further supported its arguments given that the “issues” in the two cases “substantially overlap.”¹⁵ 
   
• The General Services Administration (GSA) continues to move toward an April 2026 approval of its proposed new standard contractual language requiring AI businesses to permit “all lawful uses” of their products, among other restrictions, as discussed in last week’s article.

Conclusion and implications for AI companies contracting with the government

For AI companies contracting with defense, intelligence or other government agencies, this opinion offers some comfort that retaliatory actions may be blocked. It demonstrates that public advocacy related to AI safety and governance is protected speech, even in connection with government procurement negotiations. And it emphasizes that the government cannot penalize a domestic company as a national security risk for publicly articulating policy positions or seeking contractual provisions with which the government disagrees. The takeaway for AI companies is pragmatic: Principled advocacy remains viable, but it must be paired with deliberate risk management.

It remains the case that AI companies are operating in a politically charged environment and disagreement with agency positions can escalate quickly, resulting in immediate and severe commercial consequences. Therefore, companies should take steps to safeguard against risk, including maintaining clear documentation, seeking the advice of counsel and escalating concerns early, getting educated about applicable whistleblower protections, avoiding unapproved technical guardrails, and understanding certification and supply chain obligations.

Footnotes

1. Anthropic PBC v. U.S. Dep’t of War, Case No. 26‑cv‑01996‑RFL (N.D. Cal. Mar. 26, 2026), Opinion & Order Granting Motion for Preliminary Injunction, ECF No. 134. 

2. Id. at 31. 

3. Id. at 32.

4. See Anthropic PBC v. U.S. Dep’t of War, No. 26‑1049 (D.C. Cir.).

5] Anthropic PBC v. U.S. Dep’t of War, Case No. 26‑cv‑01996‑RFL (N.D. Cal. Mar. 26, 2026), Opinion & Order Granting Motion for Preliminary Injunction, ECF No. 134at 20-21. 

6. Id. at 24-29. 

7. Id. at 27 (emphasis in original). 

8. Id. at 29. 

9. Id. at 30-37. 

10. Id. at 30 (quoting Dkt. No. 138 at 41). 

11. Id. at 36. 

12. Id. at 31-32. 

13. Id. at 32. 

14. Id. at 1. 

15. Anthropic PBC, No. 26‑1049 (D.C. Cir.), Document No. 2165837. 

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.