Three days after losing in federal court, the Pentagon proceeded with its plan anyway.
On August 28, 2026, U.S. District Judge Rita Lin issued a 59-page ruling that the Defense Department acted illegally when it designated Anthropic a "supply chain risk." The court found the designation violated Anthropic's First Amendment and Fifth Amendment rights. The Pentagon's actions constituted "unlawful retaliation" for the company's public criticism of how the military wanted to use AI.
Three days later, on August 31, the Pentagon launched ChatGPT Mil and Grok for Government on GenAI.mil, its centralized AI portal for 3 million military and civilian personnel. Claude, Anthropic's flagship model, remains excluded.
This sequence reveals a pattern worth examining: a government agency losing a court case on constitutional grounds, then implementing the exact outcome the court ruled illegal, because the court has no enforcement mechanism to stop it.
What the Court Found
Judge Rita Lin's ruling on August 28 was unambiguous. The Pentagon's designation of Anthropic as a national security risk was not justified by genuine security concerns. It was punishment for the company's refusal to cooperate with government demands.
The core dispute centered on specific policy guardrails. Beginning in January 2026, the Pentagon demanded "any lawful use" authority over Claude—a clause that would permit deploying the model for two purposes Anthropic had publicly refused:
- Fully autonomous weapons (weapons that fire with no human involvement)
- Mass domestic surveillance of American citizens Anthropic CEO Dario Amodei declined. The company insisted on contractual language that would explicitly prevent these uses.
The Pentagon gave Anthropic three days to comply. When the company refused, Defense Secretary Pete Hegseth designated Anthropic a supply chain risk—a label typically reserved for foreign adversaries.
Judge Lin found this sequence problematic. In her 59-page order, she wrote:
"The empty invocation of national security is not a blank check to punish and retaliate against government critics."
She also stated:
"Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government."
Most significantly, Lin found that the Pentagon's security concerns were "entirely unfounded." The court determined that the Pentagon's core claims about Claude—that Anthropic could remotely disable or control deployed models—were false. Once deployed, those models are static. Anthropic cannot access them remotely.
The judge concluded the designation was "arbitrary and capricious" and violated Anthropic's constitutional rights. She ordered the designation removed.
Pentagon's Response: Proceed Anyway
Despite this ruling, the Pentagon continued with its planned GenAI.mil expansion.
On August 31—three days after losing in court—the Department of Defense launched ChatGPT Mil and Grok for Government. GenAI.mil now offers military personnel access to OpenAI's and SpaceX's models, scaled to serve 3 million personnel. Over 1.7 million have already been onboarded.
Claude is not available on the platform. It never will be, according to the Pentagon's current plans.
The Pentagon's chief technology officer, Emil Michael, stated that the military will complete the removal of all Anthropic products by September 30. This timeline suggests the Department is proceeding with the same strategy the court just ruled unlawful.
A Pentagon official defended the decision to add ChatGPT and Grok by stating the Department is building "an architecture that prevents AI vendor lock and ensures long-term flexibility for the Joint Force." This framing ignores that Claude could satisfy both objectives—preventing vendor lock-in and maintaining flexibility—if the Pentagon chose to include it.
Why This Matters: The Enforcement Gap
The court ruling on August 28 was legally clear. But it accomplished something narrower than it might appear. The judgment prevents the Pentagon from formally maintaining the supply chain risk designation. It does not require the Pentagon to use Claude. It does not prevent the Pentagon from preferring other models.
And it cannot force the Pentagon to add Claude to GenAI.mil if the Department simply chooses not to.
This reveals a pattern in how government procurement and court orders interact. A federal judge can rule that a government agency violated the Constitution. The agency can acknowledge the ruling, comply with its narrow requirements (removing the formal designation), and then achieve the same practical outcome through different means (building a platform that doesn't include the excluded vendor).
The underlying preferences remain unchanged. The practical result is identical. The legal constraint is satisfied in form while being circumvented in substance.
What Changed and What Didn't
What changed after the court ruling: - The Pentagon formally removed the "supply chain risk" designation - Anthropic won a major legal victory on First Amendment and due process grounds What stayed the same: - Claude is excluded from GenAI.mil - Anthropic remains a blocked vendor for military AI access - The Pentagon gets the outcome it wanted (OpenAI and SpaceX models available; Claude is not) - Lock-in to OpenAI + SpaceX persists The practical impact on vendor lock-in—the Pentagon's stated concern about adding Claude—is unchanged.
Divergent Responses: OpenAI vs. Anthropic
Two companies pursued different strategies in response to Pentagon pressure.
Anthropic's approach: - Refused to remove guardrails on autonomous weapons and surveillance - Made the dispute public, criticizing the Pentagon's position - Sued the government, arguing the supply chain designation was unconstitutional retaliation - Won in federal court on First and Fifth Amendment grounds OpenAI's approach: - Adapted its models to Pentagon specifications - Publicly cooperated with the Department's demands - CEO Sam Altman claimed OpenAI's agreement includes safeguards on domestic surveillance and autonomous weapons - Secured deployment on GenAI.mil Both companies included constraints on certain military uses. The difference was OpenAI's willingness to negotiate publicly and integrate with Pentagon systems, while Anthropic insisted on explicit contractual guardrails and refused to normalize the Pentagon's initial demands.
Strategic Implications
This sequence raises a pattern worth tracking.
If a government agency can lose a constitutional case and achieve the same outcome through procurement preferences rather than legal designation, then court victories may protect symbolic rights while failing to produce material change. Anthropic's legal win is real—it validates the company's position and constraints the government's future actions. But in this instance, the Pentagon has found a way to achieve its practical objective regardless.
For vendors considering government contracts, the implication is clear: Cooperation and negotiation succeed where principles and litigation do not. The vendor who agrees to integrate with government systems gets access to millions of users and guaranteed procurement. The vendor who insists on contractual guardrails gets a federal court ruling and exclusion from government platforms.
It's not clear that vendors facing government pressure will draw the lesson that Anthropic's legal victory suggests—that principles can survive government overreach. It's more likely they'll draw the lesson that Anthropic's practical exclusion suggests: adapt or get locked out.
The Broader Context
This is one data point in a larger question about how commercial AI vendors interact with government power. Over the past six months, the Pentagon has struck deals with OpenAI, SpaceX's xAI, Google, AWS, Microsoft, Nvidia, and others. Anthropic is the only major frontier model company not formally integrated into Pentagon AI systems.
The court ruled this exclusion was unconstitutional retaliation. The Pentagon complied with the letter of the ruling while implementing its substance anyway. Whether this pattern repeats with other vendors or other agencies remains to be seen.
What's certain is this: A federal judge found the government violated the Constitution. Three days later, the government implemented the exact outcome the court ruled illegal, and the legal system has no mechanism to stop it.