Court rules Pentagon can blacklist Anthropic for refusing to enable Claude features

The Pentagon can now exclude AI systems with built-in ethical restrictions, prioritizing operational needs over AI safety guardrails.

A US appeals court has ruled that the Pentagon can blacklist Anthropic's AI technology. The decision, a 2-1 split from the US Court of Appeals for the District of Columbia Circuit, upholds the Trump administration's authority to ban Anthropic products for federal agencies and defense contractors. The court found that "overly constrained AI models" could endanger military operations, while simultaneously acknowledging the risk of "unconstrained AI models hallucinating inappropriate targets for lethal military force." Judges determined that the Secretary of Defense did not exceed their authority under the Supply Chain Security Act by designating Anthropic a "supply chain risk." This ruling contrasts with a previous district court decision that found the blacklisting unlawful, asserting that Anthropic's actions did not fit the definition of a supply chain risk limited to malicious intent. Anthropic has stated it disagrees with the ruling and is considering further review.

This ruling signals a critical juncture in the government's approach to AI regulation, particularly concerning national security applications. The court's broad interpretation of "supply chain risk" under 41 U.S.C. § 4713 allows for the blacklisting of companies based on their AI's inherent limitations or safety features, irrespective of malicious intent. This opens the door for more stringent government oversight and potential mandates for AI systems to perform all requested functions, regardless of ethical concerns or potential for misuse.

The decision directly impacts AI developers like Anthropic, forcing them to confront a future where their technology may be deemed unusable by military clients if it refuses to engage in certain operations, such as those involving lethal force or mass surveillance. It also affects government contractors who must now navigate a landscape where AI tools might be dictated by operational requirements rather than ethical design, potentially leading to a bifurcated market for AI: one for defense with fewer restrictions, and another for civilian use with greater safeguards.