Pentagon Flags Anthropic’s AI Safety Limits as Unacceptable Wartime Risk, Cites

The Pentagon’s objection hinges on a 40‑page filing submitted to a California federal court, where the Department of Defense argued that Anthropic’s policy of refusing “all lawful uses” of its Claude model creates an “unacceptable” wartime risk for U.S. national‑security systems. According to Forbes, the filing contends that the company’s safety guardrails effectively limit the government’s ability to deploy the model in combat‑related scenarios, a restriction the DOD says could cripple mission‑critical AI integration. The filing also labels Anthropic a “supply‑chain risk,” echoing a separate notification the Pentagon sent to the firm, as reported by MSN.

Anthropic responded on Friday with two sworn declarations, asserting that the Pentagon’s case rests on technical misunderstandings. TechCrunch notes that the company argues the government’s concerns were never raised during months of negotiations and that the two sides were “nearly aligned” before the filing. Anthropic’s declarations claim the DOD mischaracterized the scope of its safety limits, insisting that the restrictions are designed to prevent misuse rather than impede legitimate military applications. The company further maintains that its approach aligns with industry‑wide best practices for responsible AI deployment.

The dispute arrives amid heightened scrutiny of AI supply‑chain security, especially as adversaries exploit generative models for disinformation and cyber‑operations. Wired’s recent coverage of the Predatory Sparrow hacker group, which has been targeting Iran’s financial infrastructure, underscores the broader geopolitical stakes of AI misuse. While the article does not directly reference Anthropic, the parallel illustrates why the Pentagon is wary of any AI system that cannot be fully leveraged for defensive or offensive purposes in a wartime context.

Analysts view the Pentagon’s stance as a test case for how the U.S. government will manage AI vendor risk moving forward. If the DOD’s filing succeeds, it could set a precedent that forces AI firms to relax safety constraints for government contracts, potentially reshaping the balance between ethical safeguards and national‑security imperatives. Conversely, Anthropic’s pushback may reinforce the industry’s argument that robust safety controls are essential to prevent unintended escalation or collateral damage, even in high‑stakes environments. The outcome will likely influence future procurement policies and could dictate whether AI providers are classified as “critical” or “restricted” components of the defense supply chain.