A reported ultimatum from U.S. Defense Secretary Pete Hegseth to AI startup Anthropic — demand full Pentagon access to its Claude model or face exclusion from defense contracts or compelled cooperation under the Defense Production Act (DPA) — has sharpened tensions over control of advanced AI.
According to people familiar with the discussions, Hegseth summoned Anthropic CEO Dario Amodei to Washington and asked for unrestricted Department of Defense use of Anthropic’s generative models. Anthropic confirmed the meeting, saying Amodei thanked the department for its work and that the company remains engaged in “good-faith conversations about our usage policy to ensure Anthropic can continue to support the government’s national security mission in line with what our models can reliably and responsibly do.”
The Financial Times reported that Hegseth delivered two explicit threats: to bar Anthropic from Pentagon supply chains and to invoke the DPA, a Cold War–era authority that can direct domestic industry and prioritize government orders. A senior Pentagon official told the FT the DPA could be used to force Anthropic’s cooperation “regardless of if they want to or not.”
Anthropic has so far resisted granting blanket access for classified military tasks, particularly those that might involve autonomous targeting decisions or domestic mass surveillance. The company has built a public reputation as safety-first and is understood to view some military uses as incompatible with its stated safety guardrails and long-term goals.
Geoffrey Gertz, a senior fellow at the Center for a New American Security, warned that invoking the DPA against an AI firm would be unprecedented and could undermine Anthropic’s capacity to lead on responsible AI. He cautioned that heavy-handed government action that limits a company’s markets risks undercutting broader policy aims around safe AI development.
Anthropic has already supplied its Claude model to U.S. intelligence and defense agencies since November 2024 and holds a $200 million Department of Defense contract awarded in July 2025 to prototype frontier AI capabilities for national security. The Wall Street Journal reported that U.S. forces used Claude during the 2026 raid that captured Venezuela’s Nicolás Maduro; neither Anthropic nor the Defense Department has publicly detailed that claim.
The company emphasizes careful deployment in government work, stressing that reliability, interpretability and steerability are critical when AI is used in sensitive contexts. Founded in 2021 by former OpenAI staff and led by Amodei, Anthropic says its mission is to make AI a force for human progress rather than peril.
Yet on Feb. 24 Anthropic announced it would relax parts of its core safety policy to stay competitive with other leading models. The company said the policy landscape has shifted toward prioritizing AI competitiveness and growth while federal safety rules remain underdeveloped; a spokesperson added the change was not related to Pentagon talks. The move comes amid intense rivalry with OpenAI, Google and others and against the backdrop of limited U.S. AI regulation.
The standoff raises ethical and strategic questions: can Anthropic preserve a safety-first posture if it accedes to Pentagon demands, or if the government uses the DPA to compel its technology into controversial military or surveillance roles? It also reflects a broader tilt toward a more interventionist industrial policy under the Trump administration, which has already made large direct investments in strategic sectors — from a multibillion-dollar stake in Intel to support for rare-earth suppliers — signaling a willingness to intervene where national interests are deemed critical.
Industry leaders and CEOs now face a landscape where government influence on strategic corporate choices appears more direct. Critics say this departs from America’s historically hands-off stance toward tech development; supporters argue that stronger government leverage is necessary to secure defense priorities.
Edited by: Ashutosh Pandey