美国上诉法院维持将 Anthropic 列为供应链风险的认定

华盛顿特区一个联邦上诉法院合议庭周五维持了五角大楼对 Anthropic 的列入黑名单决定，这对这家人工智能公司在与特朗普政府持续数月的争端中构成打击。

这项 2 比 1 的裁决驳回了 Anthropic 的主张。Anthropic 认为，国防部对其 Claude 模型的禁令武断、未经授权且违宪。

“国防部有充分依据得出结论：无论由国防部还是其承包商继续将 Claude 整合进国防部的信息系统，都会带来法律所涵盖的国家安全风险。”美国哥伦比亚特区巡回上诉法院法官 Gregory Katsas 在多数意见中写道，法官 Neomi Rao 加入了该意见。Katsas 和 Rao 均由总统唐纳德·特朗普任命。

由前总统乔治·H·W·布什任命的法官 Karen LeCraft Henderson 持不同意见。

今年 3 月，在关于军方如何使用其 Claude AI 模型的谈判失控后，国防部将 Anthropic 列为供应链风险，意味着该公司被指可能威胁美国国家安全。这项认定禁止美军使用 Anthropic 的模型，也禁止国防承包商在为该机构开展工作时使用这些模型。

自那以后，Anthropic 与特朗普政府的关系一直紧张，特朗普还多次在社交媒体上抨击该公司 CEO Dario Amodei。Amodei 最近因呼吁全行业放缓发展而引发特朗普不满；周四白宫为中国国家主席习近平举办的豪华国宴也没有邀请他。

特朗普周一在 Truth Social发文称：“特朗普政府已经阻止 AI‘人士’做坏事或可能做坏事，例如 Dario（Anthropic！）；他现在正假装成一个‘完美的小天使’——而我们将继续这样做！”

Anthropic 于 3 月在旧金山美国联邦地区法院和哥伦比亚特区巡回上诉法院起诉特朗普政府，寻求推翻对其的列入黑名单决定。国防部援引两项不同认定来支持其供应链风险行动，因此相关争议必须分别在两家法院审理。

旧金山一名联邦法官上个月裁定，其中一项认定违法。哥伦比亚特区上诉法院周五的裁决则维持了另一项认定。

Anthropic 一名发言人在声明中向 CNBC 表示：“我们尊重地不同意法院的裁决。另一家联邦法院已经裁定政府的平行认定不合法。我们仍然对自身立场充满信心，并正在考虑所有选择，包括寻求进一步审查。”

上诉合议庭表示，将暂缓该裁决的即时生效，以便 Anthropic 有时间向同一合议庭申请重新审理，或请求由哥伦比亚特区巡回上诉法院全体法官进行全院复审。Anthropic 也可以请求最高法院受理此案。

2026 年 5 月 31 日，弗吉尼亚州阿灵顿，美国国防部总部五角大楼的航拍画面。

Daniel Slim | Afp | Getty Images

在双方冲突于今年早些时候进入公众视野之前，Anthropic 曾是多个美国政府机构——包括国防部——的早期合作伙伴。Anthropic 于 2025 年 7 月与五角大楼签署了一份价值 2 亿美元的合同；但当该公司于当年 9 月开始就 Claude 部署到国防部 GenAI.mil AI 平台进行谈判时，谈判破裂。

国防部希望 Anthropic 允许军方在所有合法用途下不受限制地访问其模型，而 Anthropic 则希望获得保证：其技术不会被用于全自主武器或国内大规模监控。双方未能达成协议，国防部长 Pete Hegseth 指责 Anthropic 试图“攫取对美国军队作战决策的否决权”。

Katsas 周五表示，Hegseth 提出了一个“极其令人警醒”的观点，即“受到过度限制”的 AI 模型可能会意外关闭，同时 Claude 也可能“受到操纵”。尽管 Anthropic 反驳了这些说法，Katsas 表示，决策权最终属于特朗普和 Hegseth。

Katsas 写道：“在我们的共和国中，必须由总统和战争部长决定如何最好地平衡相互竞争的风险。在本案中作出这一决定时，部长并未逾越《供应链安全法》或宪法对其权力设定的任何限制。”

观看：OpenAI、Anthropic 和 Hugging Face 的 CEO 将向联合国安理会作简报

A federal appeals court panel in Washington, D.C., on Friday upheld the Pentagon's blacklisting of Anthropic, dealing a blow to the artificial intelligence company in its months-long battle with the Trump administration.

The 2-1 decision rejected Anthropic's argument that the Department of Defense's ban on its Claude models was arbitrary, unauthorized and unconstitutional.

"The Department had ample support for its conclusion that the continued integration of Claude into the Department's information systems, by the Department or its contractors, presented a statutorily covered national-security risk," Judge Gregory Katsas wrote in the majority opinion for the U.S. Court of Appeals for the District of Columbia, which Judge Neomi Rao joined. Katsas and Rao were appointed by President Donald Trump.

Judge Karen LeCraft Henderson, who was appointed by former President George H.W. Bush, dissented.

In March, the DOD labeled Anthropic a supply chain risk, meaning the company purportedly threatened U.S. national security, after negotiations about how the military could use its Claude AI models spiraled out of control. The designation prevents the U.S. military from using Anthropic's models and blocks defense contractors from using them in their work with the agency.

Anthropic's relationship with the Trump administration has been fraught ever since, and Trump has repeatedly slammed company's CEO Dario Amodei on social media. Amodei recently drew Trump's ire by calling for an industry wide slowdown, and he was not invited to the glitzy state dinner the White House hosted for Chinese President Xi Jinping on Thursday.

"The Trump Administration has stopped AI "people" from doing bad, or potentially bad, "things," like Dario (Anthropic!), who is now pretending to be a "perfect little angel" - and we will continue to do so!," Trump wrote in a post on Truth Social on Monday.

Anthropic sued the Trump administration in U.S. District Court in San Francisco and in the D.C. Circuit Appeals Court in March, seeking to reverse its blacklisting. The DOD relied on two distinct designations to justify its supply chain risk action, which meant they had to be litigated in two separate courts.

A San Francisco federal judge ruled last month that one designation was illegal. The ruling Friday by the D.C. appeals court upheld the second designation.

"We respectfully disagree with the court's decision," an Anthropic spokesperson told CNBC in a statement. "Another federal court has already held the government's parallel designation unlawful. We remain confident in our position and are considering all options, including further review."

The appellate panel said it would delay the decision from taking immediate effect to give Anthropic time to petition the same panel for a rehearing or to seek an en banc rehearing of the case, by all of the judges on the D.C. Circuit Court of Appeals. Anthropic could also ask the Supreme Court to take the case.

An aerial view of the Pentagon, which houses the US Department of Defense headquarters, in Arlington, Virginia, on May 31, 2026.

Daniel Slim | Afp | Getty Images

Before the clash between the two sides spilled into the public's view earlier this year, Anthropic served as an early partner across many U.S. agencies, including the DOD. Anthropic signed a $200 million contract with the Pentagon in July of 2025, but as the company began negotiating Claude's deployment on the DOD's GenAI.mil AI platform that September, talks collapsed.

The DOD wanted Anthropic to grant the military unfettered access to its models across all lawful purposes, while Anthropic wanted assurance that its technology would not be used for fully autonomous weapons or domestic mass surveillance. They were unable to come to an agreement, and Defense Secretary Pete Hegseth accused Anthropic of attempting to "to seize veto power over the operational decisions of the United States military."

Katsas said Friday that Hegseth raised the "deeply sobering" idea that "overly constrained" AI models could shut down unexpectedly, as well as the potential that Claude might be "subject to manipulation." While Anthropic refuted those claims, Katsas said decision-making authority ultimately rests with Trump and Hegseth.

"In our Republic, it is the President and the Secretary of War who must determine how best to balance the competing risks," Katsas wrote. "In doing so here, the Secretary did not transgress any limits on his authority under the Supply Chain Security Act or the Constitution."

WATCH: CEOs of OpenAI, Anthropic and Hugging Face to brief UN Security Council