Court Affirms Pentagon’s Right to Blacklist Anthropic Over Claude Feature Dispute

A recent ruling from a U.S. appeals court has significant implications for Anthropic, the AI company behind the model known as Claude. The court’s decision emphasized that the definition of “any person” in a relevant regulation is not limited solely to adversarial entities or foreign powers. This finding suggests that concerns over potential obstructions by Anthropic could impact national security operations.

According to the court, the use of the term “deny” indicates that Anthropic might restrict the use of certain functionalities within the Claude system. The majority opinion of the court concluded, “We determine that the Secretary’s concern regarding the possibility of Anthropic disabling Claude’s lawful actions denotes a ‘supply chain risk’ as defined by section 4713.” The judges expressed that there are reasonable fears surrounding Anthropic potentially altering Claude’s capabilities to limit functions deemed essential for national security by the Department of Defense.

Dissenting Opinion

Judge Karen Henderson, appointed during the George H.W. Bush administration, presented a dissenting viewpoint. She challenged the majority’s interpretation of section 4713, arguing that the statutory language pertains primarily to deliberate attempts to hinder or surveil the operation of any covered technology within federal systems. Henderson underscored that Congress passed the statute in response to growing concerns about threats from hostile nation-states and other malicious actors attempting to infiltrate government technology systems.

Her dissent stressed that the statute’s intent should not extend to the enforcement of restrictions put in place by contractors, which may be necessary to uphold legal or ethical standards against government wishes.

Faced with governmental pressures, Anthropic has claimed that the Trump administration engaged in illegal retaliation after the firm refused to relax its product restrictions against lethal autonomous weaponry and mass surveillance operations targeting civilians. A district court in California initially supported this claim, highlighting potential implications for corporate governance in the technology sector.

The appeals court decided that Anthropic’s restrictions embedded within Claude prevented the model from executing specific tasks requested by users within the government. Notably, disputes have arisen around contractual limitations that could affect the model’s application during military operations, contributing to uncertainty regarding its operational effectiveness.

Judge Rita Lin, who presided over the case in the Northern District of California, ruled that the government’s actions constituted a First Amendment violation. She argued that vague national security claims should not serve as a pretext for penalizing dissenting voices within the technology sector.