Trump Invokes Defense Production Act in Escalating War with Anthropic

The escalating confrontation between the Trump administration and Anthropic represents a watershed moment in the relationship between the federal government and the artificial intelligence industry. At the heart of the dispute is a fundamental paradox in executive policy: the administration has labeled Anthropic a national security risk while simultaneously asserting that its technology is so vital to the national interest that the government must compel its delivery without safety restrictions. This dual-track strategy—utilizing both the threat of blacklisting and the coercive power of the Defense Production Act (DPA)—signals a new era where software and algorithmic weights are treated with the same strategic gravity as physical munitions or raw materials during the Cold War.

The catalyst for this aggressive stance appears to be the recent release of Claude Code, a suite of developer tools that demonstrated such significant leaps in autonomous coding capability that it triggered a massive $1 trillion sell-off across the global software sector. For the administration, this market volatility served as proof of Anthropic’s disruptive potential, leading to a swift reclassification of the company as a supply chain risk. By threatening to bar any company that does business with Anthropic from receiving defense contracts, the administration is effectively attempting to isolate the firm from the broader commercial ecosystem, a move that analysts suggest could be a 'death blow' for the venture-backed startup.

“

Anthropic CEO Dario Amodei has responded with a firm refusal to comply, stating that the company cannot 'in good conscience' accede to demands that would strip away the safety measures defining their product.

However, the administration’s objectives extend beyond mere containment. Defense Secretary Pete Hegseth and President Trump have signaled through social media that they intend to use the DPA to force Anthropic to provide the Claude model 'without caveats.' This phrasing is a direct challenge to Anthropic’s 'Constitutional AI' framework, which embeds safety guardrails and ethical constraints directly into the model's training process. The administration’s demand for a raw, unrestricted version of Claude suggests a desire to weaponize the AI for offensive cyber operations or autonomous systems that might otherwise be blocked by the company’s internal safety protocols. This creates a direct ideological and legal collision between the state’s perceived military requirements and a private corporation’s ethical commitments.

Anthropic CEO Dario Amodei has responded with a firm refusal to comply, stating that the company cannot 'in good conscience' accede to demands that would strip away the safety measures defining their product. The company is now preparing for a protracted legal battle in the federal courts, challenging the government's authority to use the DPA to seize intellectual property and force the creation of potentially dangerous software variants. This case will likely test the limits of executive power in the digital age, specifically whether the DPA—originally designed to ensure the production of steel and rubber—can be legally applied to the intangible and highly sensitive weights of a large language model.

For the broader defense-tech sector, the implications are chilling. The administration's willingness to use tweets as the primary medium for communicating major policy shifts and DPA invocations introduces a level of unpredictability that could stifle investment in AI safety and alignment. If the government succeeds in forcing Anthropic’s hand, it sets a precedent that no AI developer can truly own or control the safety parameters of their models if the state deems them strategically necessary. Investors and competitors are now watching closely to see if this 'war' on Anthropic is an isolated incident or the beginning of a broader campaign to nationalize or strictly regulate the most advanced tiers of American artificial intelligence.