在2026年3月10日格林威治时间+8 9:05，五角大楼研究与工程负责次官 Emil Michael 表示，与 Anthropic PBC 重新恢复用于军事用途的 AI 合作几乎无望。Anthropic 在周一提起诉讼，试图阻止国防部及其他机构将其列为“供应链风险”（supply-chain risk）并禁止其参与政府合同，这一诉讼被 Pentagone 方视为“预期反应”；Michael 说，即使诉讼成功也不会改变军方决定。Anthropic 发言人未对采访即时报以回应。
在与 Anthropic 首席执行官 Dario Amodei 进行数周谈判后，约两周前协议因公司要求在合同中加入两项限制（不得用于美国大规模监控与自主武器部署）而中断。此后，五角大楼依据此前对“对手国家”企业才会采用的惯例，将总部位于旧金山的 Anthropic 列入供应链风险。此前 Anthropic 曾是能够在五角大楼机密云环境中运行的唯一 AI 系统，其 Claude Gov 也因易用性而受到国防人员青睐。此次决定使约2亿美元（200 million USD）的分类 AI 工具合同面临风险，并可能阻断其与其他承包商在军事领域的合作；Donald Trump 政府与五角大楼部长 Pete Hegseth 已明确给出六个月窗口，以便军方及美国机构转向其他 AI 供应商。
Anthropic 在法庭材料中称，联邦承包体系中的连带效应已出现：一名合作方警告可能暂停工作或从现有部署中移除 Claude，其他承包商也在提高警觉、暂停协作并考虑终止合同。Michael 强调 Anthropic 公开主张安全限制，反映其可能想干预作战指挥链与实际操作控制；Anthropic 则否认该点，并坚持军事决策权应由五角大楼掌握。Amodei 仍表示愿意与军方继续合作，但必须将上述两项使用边界写入合同。被问及是否会重开谈判时，Michael 直接回应：“谈判结束了，我们已经继续前进。”
On March 10, 2026 at 9:05 AM GMT+8, Pentagon Under Secretary for Research and Engineering Emil Michael said there is little chance to resume military use negotiations with Anthropic PBC. Anthropic had filed suit on Monday to block the Pentagon and other agencies from designating it as a “supply-chain risk” and excluding it from government contracts. Michael called that lawsuit an expected reaction and said the move would not change Pentagon policy; he said there is no scenario in which the matter would be resolved that way. Anthropic declined immediate comment on Michael’s remarks.
After weeks of talks between Michael and Anthropic CEO Dario Amodei, negotiations collapsed roughly two weeks ago when Anthropic reportedly demanded contractual guarantees that its AI would not be used for U.S. mass surveillance or autonomous weapons deployment. The Pentagon then labeled the San Francisco-based firm a supply-chain risk, a designation normally reserved for firms from adversarial states. Until then, Anthropic had been the only AI system approved to operate in the Pentagon’s classified cloud, and Claude Gov was widely used for its ease of use. The designation puts at risk about US$200 million in classified AI work and could block Anthropic from partnering with other defense contractors; the Trump administration and Defense Secretary Pete Hegseth reportedly set a six-month timeline for agencies to shift to other AI providers.
In court filings, Anthropic said the decision had already disrupted its federal partnerships: one vendor warned it might suspend work or remove Claude from existing deployments, while others are pausing collaborations and considering contract terminations. Michael repeated a Pentagon concern that Anthropic’s public position on safeguards suggested a desire to influence the chain of command and operational control. Anthropic denied that interpretation and said all military decisions stay with the Pentagon; Amodei said it wants to keep working with the military only if those two usage limits are honored. Asked about any return to talks, Michael answered bluntly, “The talks are over. We’re moving on.”