U.S. District Court Judge Rita Lin on Thursday declared the Pentagon’s supply‑chain risk label on AI firm Anthropic “arbitrary and capricious,” overturning an effort by the administration to bar contractors from using the company’s Claude model for certain defense work.
Judge finds retaliation behind the risk label
In her opinion, Judge Lin said the record showed the agency acted to punish Anthropic for refusing to relax safety guardrails the Department of Defense wanted removed. She wrote that the move violated the First Amendment and denied the company the pre‑deprivation process required by the Fifth Amendment.
The ruling notes that, days before the designation was issued, the secretary of defense proposed invoking the Defense Production Act to treat Anthropic as essential to national security. That same week, officials discussed collaborating with the firm on its upcoming Mythos model for sensitive projects. The judge called the contrast “inconsistent with a genuine fear that Anthropic would sabotage its software.”
She also pointed out that the usage policy governing Pentagon contracts is merely a contractual limit. “Anthropic is incapable of enforcing it technologically, and does not have direct visibility into how the department uses its model,” the opinion said. The judge found no evidence in the administrative record describing any “backdoors” or technical means that would let Anthropic disable or alter Claude during a government operation.
Related: Apple ignores decline of cheap smartphone market
Anthropic submitted unrebutted evidence that it lacks any capability to access or control deployed models. The court concluded that the agency’s stated concerns “made no sense” without a concrete technical rationale.
Legal analysts expect the administration to appeal, possibly taking the case to the Supreme Court.
Implications for technology leaders
The decision raises questions for chief information officers who rely on vendor contracts to manage risk. Alan Webber, program VP for national security, defense, and intelligence at IDC, noted that the ruling shows contractual guardrails can become a trigger for blacklisting, a scenario risk registers are meant to prevent.
Webber added that competing AI vendors have been using the government action as a sales tool, and with this ruling, the argument that Anthropic is a designated supply‑chain risk “just got weaker,” which could lead to contract award disputes.
Related: Startups Pursue Next Generation AI Models
In practice, CIOs are urged to evaluate products on merit—considering effectiveness, cost, security, safety and confidentiality—rather than relying on a blanket supply‑chain risk tag.
One analyst explained that the court rejected the leap from a policy disagreement to a broad risk designation because the agency could not demonstrate a technical basis. He emphasized that as AI becomes embedded in enterprise operations, “We don’t trust the vendor” cannot replace a defined risk model.
Organizations must be able to articulate the actual technical risk, how it manifests, what controls exist, and whether the response is proportional. Conflating model behavior, usage policies, ethics, contractual restrictions and cybersecurity into a single “AI risk” category can obscure real threats.
For now, the ruling restores Anthropic’s ability to compete for defense contracts, though the appeal process remains open.
