Anthropic vs. The Pentagon: When AI Companies Become 'Supply Chain Risks'

On February 14, 2026 — Valentine's Day, in a timing choice that felt either tone-deaf or deliberate — the Department of Defense's Office of the Under Secretary for Acquisition and Sustainment quietly published Federal Register Notice 2026-03412. Buried in procurement language that would put most readers to sleep, it contained a bombshell: Anthropic PBC, the San Francisco-based AI company behind the Claude family of models, had been designated a "supply chain risk" under DoD Directive 5200.44.

The designation effectively blacklists Anthropic from new federal contracts across all defense and intelligence agencies. Existing contracts — including a $127 million cloud AI services agreement with the Defense Intelligence Agency signed in late 2025 — are under mandatory 90-day review for potential termination.

Anthropic's response was swift and uncharacteristic. A company known for measured, academic-toned public communications filed suit in the U.S. Court of Federal Claims within 72 hours, calling the designation "arbitrary, retaliatory, and unconstitutional." CEO Dario Amodei broke his usual restraint in a blog post titled "We Will Not Be Coerced," writing: "A government that punishes companies for taking safety seriously is a government that will get the AI it deserves — fast, cheap, and dangerous."

The hearing is set for March 24, 2026. The stakes extend far beyond one company.

The Designation: What Actually Happened

To understand why this matters, you need to understand what a "supply chain risk" designation is — and what it is not.

Under the Federal Acquisition Supply Chain Security Act (FASCSA) of 2018 and subsequent DoD implementation directives, the government can designate companies as supply chain risks if their products or services pose threats to the integrity, security, or resilience of federal information systems. The authorities were designed primarily to address foreign adversary threats — the legislation was drafted in the shadow of the Huawei controversy and concerns about Chinese-manufactured telecommunications equipment containing backdoors.

The criteria the DoD cited in Anthropic's designation, according to the Federal Register notice and subsequent court filings, fall into three categories:

The weakness of these justifications is striking. Foreign investment? Google accepted $3.2 billion from Saudi Arabia's PIF in 2024 for cloud infrastructure without triggering any designation. Public Benefit Corporation status? Patagonia, Kickstarter, and dozens of government contractors operate as PBCs. Model availability restrictions? Every AI company has acceptable use policies that limit certain applications.

What makes Anthropic different is not what's in the Federal Register notice. It's what isn't.

The Real Story: Safety as Insubordination

Sources familiar with the internal Pentagon deliberations — speaking on condition of anonymity because the discussions involved classified procurement assessments — describe a designation driven less by genuine supply chain security concerns and more by institutional frustration with Anthropic's posture toward military applications.

The timeline is revealing:

• August 2025: Anthropic publishes updated Responsible Scaling Policy (RSP) v3.0, explicitly prohibiting Claude deployment in autonomous weapons systems and requiring human-in-the-loop oversight for any lethal force decision support.
• September 2025: DoD's Chief Digital and AI Officer (CDAO) requests that Anthropic develop a "defense-optimized" variant of Claude with relaxed safety restrictions for classified environments. Anthropic declines, offering instead to work within its existing RSP framework.
• October 2025: Pentagon procurement officials begin informal consultations with the Defense Counterintelligence and Security Agency (DCSA) about Anthropic's foreign investment profile.
• November 2025: Anthropic publicly opposes a DoD proposal to exempt AI systems used in "time-critical targeting" from existing autonomous weapons review processes.
• December 2025: DCSA completes its assessment. The supply chain risk designation process begins.
• February 2026: Designation published.

The pattern suggests a government that asked an AI company to compromise its safety commitments, was told no, and then went looking for a procurement mechanism to punish the refusal.

> "This is not a supply chain case. This is a compliance case dressed up in national security clothing." — Professor Rachel Harmon, University of Virginia School of Law, in written testimony submitted to the Senate Armed Services Committee

The Legal Battle: Constitutional Territory

Anthropic's lawsuit raises three primary claims, each with significant implications beyond this case.

Due Process (Fifth Amendment)

The supply chain risk designation was issued without prior notice to Anthropic and without an opportunity to respond before publication. FASCSA provides for an exclusion and removal process, but the statute contemplates notice and an opportunity to contest — procedural safeguards the DoD appears to have bypassed. Anthropic argues this violates the Due Process Clause.

The government's counterargument — that national security procurement decisions are entitled to broad executive discretion — runs into a problem: Anthropic is not a foreign company. The Huawei designation involved a Chinese state-linked enterprise. The TikTok divestiture order involved a Chinese parent company. The constitutional calculus is fundamentally different when the government targets a domestic company founded by American citizens, incorporated in Delaware, and headquartered in the United States.

Administrative Procedure Act

Anthropic alleges the designation was "arbitrary and capricious" under the APA because the DoD failed to articulate a rational connection between the stated risk factors and any actual supply chain threat. The company points out that none of the cited concerns — foreign minority investment, PBC status, safety policies, employee citizenship — have been applied as disqualifying factors to any other major AI or technology contractor.

First Amendment (Retaliation)

This is the most aggressive claim and the one that has attracted the most attention from industry groups. Anthropic argues that the designation was retaliatory punishment for the company's public speech — its Responsible Scaling Policy, its public opposition to autonomous weapons exemptions, and Dario Amodei's public commentary on AI safety. If successful, this argument would establish that the government cannot use procurement authority to punish companies for expressing views on how their technology should be used.

The Amicus Avalanche

The breadth of industry support for Anthropic's position has been remarkable — and revealing.

The Information Technology Industry Council (ITI), whose members include Microsoft, Google, Apple, Amazon, and Meta, filed a 42-page amicus brief arguing that the designation "threatens the foundational assumptions on which the technology industry's relationship with the federal government has operated for decades." The brief notes that U.S. technology companies hold approximately $284 billion in active federal contracts and that subjecting any company to supply chain risk designations based on corporate governance choices or public policy positions would "inject paralyzing uncertainty into the largest technology procurement market in the world."

The Computer & Communications Industry Association (CCIA) focused on procedural due process, arguing that the lack of pre-deprivation notice and hearing violates established Supreme Court precedent on government blacklisting.

Perhaps most notable was the brief filed by the AI Alliance — whose members include both Anthropic's competitors and its collaborators. The brief makes an argument that transcends competitive dynamics: if the government can designate an AI company as a supply chain risk because its safety policies are too restrictive, the rational response for every AI company is to weaken its safety policies. The brief calls this a "race-to-the-bottom dynamic that directly undermines the national security interests the designation purports to protect."

Six amicus briefs from six major organizations, all siding with Anthropic. The government has received none in support of its position.

Historical Parallels — and Why They Break Down

The instinct to compare this case to Huawei and TikTok is understandable but ultimately misleading.

Huawei (2019-2020)

The Huawei designation under Section 889 of the NDAA was grounded in specific intelligence findings about the company's relationship with the Chinese Communist Party and the People's Liberation Army. Huawei's founder, Ren Zhengfei, is a former PLA engineer. Chinese national intelligence law compels Chinese companies to cooperate with state intelligence operations. The company was incorporated in China, governed by Chinese law, and subject to CCP oversight through internal party committees.

None of this applies to Anthropic.

TikTok (2020-2025)

The TikTok saga — from Trump's initial executive order through Biden's divestiture legislation to the Supreme Court's 2025 ruling upholding the ban — centered on data sovereignty concerns tied to ByteDance's Chinese ownership and the potential for CCP access to American user data under Chinese national security law.

Again, none of this applies to Anthropic. The company's data centers are in the United States. Its models are trained on data governed by U.S. law. Its corporate governance is subject to Delaware corporate law and SEC oversight.

The Better Analogy: Blacklisting in the McCarthy Era

Legal scholars have drawn a more uncomfortable parallel. In the 1950s, the federal government used loyalty programs and procurement blacklists to punish individuals and organizations whose political views were deemed insufficiently aligned with government priorities. The Supreme Court eventually struck down many of these programs as violations of due process and free association rights.

Professor David Pozen of Columbia Law School has argued that the Anthropic designation represents "a modern procurement loyalty test — the government is effectively asking AI companies to demonstrate their loyalty to defense priorities as a condition of market access, and punishing those that prioritize other values."

The Money at Stake

The financial implications of the designation are substantial — for Anthropic and for the broader AI industry.

Anthropic's federal business represented approximately $340 million in annual contract value as of January 2026, roughly 8% of the company's projected $4.2 billion in 2026 revenue. The direct revenue loss is significant but not existential.

The indirect effects are far more damaging:

• Investor confidence: Anthropic's last primary round valued the company at $61.5 billion. At least three institutional investors have reportedly placed follow-on investments on hold pending resolution of the case.
• Talent retention: Anthropic employs approximately 1,800 people, many of whom hold security clearances tied to federal work. The designation puts those clearances — and the careers built on them — in jeopardy.
• Commercial spillover: Several Fortune 500 companies with significant defense business have reportedly paused or delayed enterprise Claude deployments, citing concerns about "guilt by association" in the procurement ecosystem.
• Global perception: Allied governments in the Five Eyes intelligence alliance are watching closely. The UK's AI Safety Institute and Australia's Signals Directorate both have active collaborations with Anthropic that could be complicated by a U.S. supply chain risk designation.

The Federal AI Procurement Market

The broader market context amplifies the stakes. Federal AI spending has grown from $3.3 billion in FY2023 to an estimated $18.7 billion in FY2026, driven by the DoD's Replicator initiative, intelligence community modernization programs, and civilian agency automation mandates.

This is one of the fastest-growing procurement categories in federal history. Any AI company locked out of this market faces a significant competitive disadvantage — not just in lost revenue, but in lost access to the data, feedback loops, and operational experience that government deployments provide.

The Industry Response: Quiet Panic

Publicly, AI companies have been measured in their responses. Privately, the industry is in turmoil.

At least two frontier AI labs — neither of which Signal is naming at this time — have reportedly revised their responsible use policies since the designation was announced, softening language around military applications and removing explicit prohibitions on autonomous weapons support. One company's revised policy replaced "We will not develop AI systems for autonomous lethal targeting" with "We will work with government partners to ensure appropriate human oversight in sensitive applications."

The semantic difference is vast. The behavioral difference is the point.

OpenAI, which has aggressively pursued military and intelligence contracts since removing its military use prohibition in January 2024, has said nothing publicly about the Anthropic designation. Google DeepMind, which maintains its own set of AI principles restricting weapons applications, has also remained silent. Meta, which open-sources its Llama models and has limited direct federal contracting exposure, declined to comment.

The silence is itself a signal. No major AI company wants to publicly defend Anthropic and risk drawing the Pentagon's attention to its own policies. No major AI company wants to publicly criticize Anthropic and validate the designation. The result is a strategic paralysis that serves the government's interests perfectly.

> "The genius of this designation, from the Pentagon's perspective, is that it doesn't need to be legally sustainable. It just needs to exist long enough to change behavior. And it's already doing that." — Former DoD acquisition official, speaking anonymously

What the March 24 Hearing Will Decide

The preliminary injunction hearing on March 24 will not resolve the underlying case, but it will answer three critical questions:

1. Does the court have jurisdiction?

The government will argue that supply chain risk designations are committed to executive discretion and are unreviewable by courts. If the court agrees, the case is over — and the executive branch will have established a procurement authority with essentially no judicial check.

2. Is Anthropic likely to succeed on the merits?

The court will evaluate the strength of Anthropic's due process, APA, and First Amendment claims. A finding of likelihood of success would signal that the designation is legally vulnerable and would put significant pressure on the DoD to negotiate.

3. Does the balance of equities favor an injunction?

The court must weigh Anthropic's irreparable harm (lost contracts, reputational damage, investor flight) against the government's interest in supply chain security. The government's challenge is that its own filing identifies no specific security incident, no data breach, no espionage allegation, and no concrete evidence of supply chain compromise — making the "irreparable harm to national security" argument difficult to sustain.

Legal experts are cautiously optimistic about Anthropic's chances. Professor Steve Vladeck of Georgetown Law has noted that "the government's strongest card — national security deference — is weakened considerably when the target is a domestic company and the alleged risks are this speculative."

The Deeper Question: Who Controls Frontier AI?

Strip away the procurement law and the constitutional arguments, and the Anthropic case is really about something more fundamental: who gets to decide how the most powerful AI systems are used?

Anthropic's position is that the companies building frontier AI have a right — and a responsibility — to set boundaries on how their technology is deployed, including in military contexts. The Pentagon's position, implicit in the designation, is that companies operating in the national security space must subordinate their own safety frameworks to government requirements.

This is not a new tension. Defense contractors have always operated under government specifications. Lockheed Martin does not get to decide which countries receive F-35s. Raytheon does not publish a "responsible use policy" for Tomahawk missiles.

But AI companies are not traditional defense contractors. They sell the same models to hospitals, schools, startups, and foreign governments that they sell to the Pentagon. Their safety frameworks are not just internal governance documents — they are product features that millions of commercial customers rely on. Weakening those frameworks for one customer weakens them for all customers.

The Anthropic case will not resolve this tension permanently. But it will establish the first legal precedent on whether the government can use procurement authority to override an AI company's safety commitments. And that precedent will shape the industry for a generation.

What Comes Next

Three scenarios:

Scenario 1: Injunction granted, designation suspended. The court blocks the designation pending full adjudication. The DoD faces political pressure to withdraw or modify the designation. Anthropic's federal business resumes. Other AI companies interpret the ruling as protection for safety-first policies. This is the best case for the industry.

Scenario 2: Injunction denied, case proceeds. The designation remains in effect during litigation that could take 12-18 months. Anthropic loses federal revenue and faces escalating indirect costs. Other AI companies accelerate their shift toward defense-friendly policies. The race to the bottom intensifies.

Scenario 3: Settlement. The most likely outcome. Anthropic agrees to modify certain RSP provisions for classified environments; the DoD withdraws the designation; both sides declare victory. The underlying legal questions remain unresolved, leaving the threat of future designations hanging over the industry.

Whatever happens on March 24, the Anthropic case has already changed the calculus for every AI company in the United States. The question is no longer whether frontier AI labs will engage with defense — it is whether they will be allowed to set any terms for that engagement.

The Pentagon has made its position clear: in the national security space, the customer sets the rules.

Anthropic is betting that the Constitution disagrees.