Anthropic-Pentagon governance dispute

Anthropic told a federal appeals court on April 22 that it cannot change Claude after the model is installed inside classified Pentagon networks. (abcnews.com) The filing, submitted to the U.S. Court of Appeals in Washington, D.C., is part of a fight over the Pentagon’s decision to treat Anthropic as a “supply chain risk.” Anthropic said the claim is wrong because the government, not the vendor, controls the deployed system in those closed networks. (abcnews.com) That dispute grew out of a canceled $200 million Pentagon contract and Anthropic’s effort to limit how Claude could be used in military systems. The company has objected to uses tied to domestic surveillance and fully autonomous weapons, while the Defense Department has said it needs authority to use artificial intelligence for “all lawful purposes.” (abcnews.com) (cbsnews.com) The legal fight now sits in two courts with different rulings. On March 26, U.S. District Judge Rita Lin in San Francisco blocked the administration from enforcing the supply-chain label, but on April 9 the D.C. appeals court refused to pause the Pentagon’s actions while the case proceeds. (cbsnews.com) (pbs.org) That split has left Anthropic in an awkward middle ground. CNBC reported the company can still work with other federal agencies, but defense contractors are barred from using Claude in Pentagon work while the Washington case continues. (cnbc.com) The case has also turned a technical contracting question into a governance one: who gets to set the rules after an AI model is sold. Anthropic says its structure is designed to keep safety commitments from being overridden by ordinary commercial pressure. (news.bloomberglaw.com) (anthropic.com) Anthropic is a public benefit corporation, and its Long-Term Benefit Trust has authority to select and remove a growing share of the board, ultimately a majority. Anthropic says the trust exists to align the company with its stated mission of developing advanced AI for the long-term benefit of humanity. (anthropic.com 1) (anthropic.com 2) Bloomberg Law argued that this design helps explain why Anthropic and OpenAI have responded differently to Pentagon business. The column says Anthropic’s trust-centered structure gives mission-focused directors more leverage when military contracts collide with safety restrictions. (news.bloomberglaw.com) The Pentagon has framed the matter differently in court and in public statements, saying Anthropic was trying to dictate military policy and that the government must retain discretion over national-security procurement. The D.C. appeals court echoed some of that deference on April 9, writing that the balance favored the government during “active military conflict.” (pbs.org) (cnbc.com) Oral arguments in the Washington appeal are scheduled for May 19. By then, the judges will be weighing not just whether Anthropic can win back Pentagon access, but whether an AI vendor can still enforce its own guardrails once its software crosses into a classified military system. (abcnews.com)