Anthropic Refuses Pentagon Demand to License AI for Autonomous Weapons

Washington, D.C. — In a landmark confrontation between artificial intelligence ethics and national defense policy, Anthropic, the San Francisco-based AI startup co-founded by former OpenAI researchers, has refused to grant the U.S. Department of Defense access to its proprietary large language models for integration into autonomous weapons systems. The decision has triggered a high-stakes response from Defense Secretary Pete Hegseth, who has publicly warned that Congress may soon pass emergency legislation to override corporate objections and mandate technology sharing under national security grounds.

According to internal documents obtained by The Decoder, the Pentagon’s Strategic Capabilities Office initiated discussions with Anthropic in early 2025 to license the company’s Claude 3.5 model for use in drone swarm coordination and target identification algorithms. Anthropic’s leadership, citing its longstanding ethical framework and public commitment to "not build weapons that make life-or-death decisions without human oversight," formally declined the request in a confidential letter dated March 12, 2025.

"We believe that autonomous lethal decision-making must never be delegated to algorithms, no matter the military justification," said Anthropic CEO Dario Amodei in a private briefing to employees. "Our mission is to build AI that benefits humanity, not one that automates war."

The Pentagon’s reaction was swift. In a televised address on April 5, Secretary Hegseth declared, "If private companies choose to withhold technology that could save American lives on the battlefield, then they will no longer be allowed to opt out. We will legislate compliance."

While the exact language of the proposed legislation remains classified, sources familiar with the draft bill — referred to internally as the "National Security AI Access Act" — suggest it would grant the Department of Defense broad authority to compel technology transfer from any U.S.-based AI firm deemed critical to defense operations. The bill would also establish a new Defense AI Review Board with subpoena power and the ability to impose fines for noncompliance.

The move has ignited a firestorm among tech ethicists, civil liberties groups, and industry peers. The Electronic Frontier Foundation called the proposed law "a dangerous precedent that turns corporate R&D into state property." Meanwhile, Google DeepMind and Microsoft’s AI division have quietly distanced themselves from the controversy, declining to comment on whether they’ve received similar Pentagon requests.

Legal scholars warn that the legislation could face constitutional challenges on grounds of corporate free speech, Fifth Amendment takings, and First Amendment protections for algorithmic expression. "This isn’t just about weapons," said Professor Elena Ramirez of Stanford Law School. "It’s about whether the government can force a private company to publish its intellectual property for military use — essentially, to weaponize its own code."

Anthropic has not ruled out legal action if the bill is enacted. The company is reportedly consulting with international partners, including the European Union’s AI Office, to explore regulatory shielding mechanisms under the EU’s AI Act, which explicitly bans autonomous lethal weapons.

As tensions escalate, the global AI community watches closely. The outcome may set a precedent for how democracies balance innovation, corporate sovereignty, and military necessity in the age of autonomous warfare. For now, Anthropic stands firm — but the clock is ticking on its right to say no.