内容来源：https://www.wired.com/story/trump-moves-to-ban-anthropic-from-the-us-government/

内容总结：

美政府与AI巨头安斯罗皮克公开决裂 国防应用伦理争议升级

当地时间周五，美国总统特朗普在社交媒体上宣布，已指示所有联邦机构“立即停止”使用人工智能公司安斯罗皮克（Anthropic）的AI工具。此举标志着美国政府与这家领先的AI实验室围绕人工智能军事化应用的长期分歧彻底公开化。

特朗普在声明中指责安斯罗皮克的“左翼激进分子”在试图“胁迫国防部”时犯下了“灾难性错误”。他同时表示，各机构将有“六个月的过渡期”来逐步停用其产品，这或为双方进一步谈判留出时间。

紧随总统表态，国防部长皮特·赫格塞斯将安斯罗皮克列为“供应链风险”。该 designation通常仅用于被认为威胁美国国家安全的外国企业，此举将禁止美军及其承包商、供应商与该AI公司合作。赫格塞斯在社交媒体上严厉批评该公司及其CEO达里奥·阿莫代，称其以“有效利他主义”为名，行“胁迫美国军方”之实，是将“硅谷意识形态置于美国人的生命之上”。

矛盾核心在于合同条款的修改。国防部试图取消去年七月与安斯罗皮克等公司协议中关于AI部署方式的限制，转而允许技术的“所有合法用途”。安斯罗皮克对此表示反对，担心修改后的条款可能允许AI用于完全控制致命性自主武器或对美国公民进行大规模监控。国防部则表示目前并未且无计划如此使用AI，但特朗普政府高级官员强烈反对由一家民用科技公司来限定如此重要技术的军事用途。

安斯罗皮克是首家与美国军方合作的大型AI实验室，去年通过与五角大楼签署的2亿美元协议，开发了限制较少的定制模型“克劳德政府版”。尽管该模型目前主要用于撰写报告、总结文件等常规任务，但也应用于情报分析和军事规划。

此次公开争执始于有报道称美军曾使用“克劳德”协助策划抓捕委内瑞拉总统的行动。此后，双方矛盾在社交媒体上持续升级。赫格塞斯本周初曾与阿莫代会面，要求其在周五前同意修改合同条款，并表达了继续合作的意愿，但最终未能达成一致。

这场风波也折射出硅谷与国防部门关系演变的深层张力。近年来，硅谷企业从普遍回避国防项目转向日益接纳并最终成为成熟的军事承包商。安斯罗皮克与五角大楼的冲突正在试探这种转变的边界。本周，数百名来自OpenAI和谷歌的员工签署公开信，声援安斯罗皮克，并批评自家公司取消AI军事应用限制的决定。

有专家认为，当前争议更多是理念层面而非具体部署上的分歧。前五角大楼新兴技术副助理部长、AI军事应用专家迈克尔·霍罗威茨指出，双方目前在实际使用案例上并无根本矛盾，争论焦点是“目前并未提上日程的理论性应用场景”。

安斯罗皮克公司秉持将安全置于核心的AI开发理念。其CEO阿莫代伊曾撰文谈及强大AI的风险，特别指出完全自主的AI控制武器虽然“在捍卫民主方面具有合法用途”，但亦是“危险武器”。

截至目前，美国国防部与安斯罗皮克均未对最新事态置评。这场围绕技术与伦理的博弈，如何影响美国AI军事化应用的未来路径，仍有待观察。

中文翻译：

美国总统唐纳德·特朗普周五宣布，已指示所有联邦机构"立即停止"使用Anthropic公司的人工智能工具。此前，Anthropic与政府高层就人工智能军事化应用问题已持续数周对峙。

特朗普在其社交平台Truth Social上发文称："Anthropic那些左翼疯子试图胁迫国防部的行为是灾难性错误。"

特朗普表示，各机构将有"六个月过渡期"逐步停用Anthropic技术，这可能为政府与这家AI初创企业进一步谈判留出时间。

五角大楼与Anthropic公司均未立即回应置评请求。

总统宣布决定后不久，国防部长皮特·赫格塞斯表示，Anthropic还将被列为"供应链风险企业"。该标签通常用于被视作威胁美国国家安全的外国企业，此举将禁止美军及其承包商、供应商与这家AI公司合作。

赫格塞斯在X平台发文，严厉抨击Anthropic及其首席执行官达里奥·阿莫代伊拒绝接受政府要求的行为："他们披着'有效利他主义'的虚伪外衣，企图胁迫美国军方屈服——这种将硅谷意识形态置于美国民众生命之上的企业道德表演，是懦夫行径。"

国防部一直试图修改去年七月与Anthropic等公司达成的协议条款，要求取消AI部署限制，转而允许对该技术进行"所有合法用途"的应用。Anthropic反对此项修改，认为这可能使AI被用于完全控制致命性自主武器或对美国公民进行大规模监控。

五角大楼目前并未以上述方式使用AI，并表示暂无相关计划。但特朗普政府高级官员已多次表态，反对由民用科技公司来主导如此重要技术的军事化应用。

Anthropic是首家与美军合作的大型AI实验室，去年通过与五角大楼签署的2亿美元协议展开合作。该公司开发了多个名为Claude Gov的定制模型，其使用限制少于常规模型。谷歌、OpenAI和xAI同期也签署了类似协议，但Anthropic是目前唯一获准接触机密系统的AI公司。

Anthropic的模型通过Palantir平台和亚马逊机密军事云平台提供服务。据一位要求匿名的知情人士向《连线》杂志透露，Claude Gov目前主要用于撰写报告、文件摘要等常规任务，但也应用于情报分析和军事规划。

近年来，硅谷企业经历了从普遍回避国防项目，到逐渐接纳并最终成为全面军事承包商的转变。Anthropic与五角大楼的争端正在考验这一转型的边界。本周，OpenAI和谷歌的数百名员工签署公开信支持Anthropic，批评自家公司取消AI军事化应用限制的决定。

据《华尔街日报》报道，OpenAI首席执行官萨姆·奥尔特曼在今日发给员工的备忘录中表示，公司赞同Anthropic的立场，同样将大规模监控和全自主武器视作"不可逾越的红线"。奥尔特曼补充说，公司将尝试与五角大楼达成允许其继续军事合作的协议。

五角大楼与Anthropic的公开争执始于Axios的报道——该报道披露美军高层曾使用Claude协助策划抓捕委内瑞拉总统尼古拉斯·马杜罗的行动。行动结束后，Palantir一名员工向美军高层转达了Anthropic员工对其模型使用方式的担忧。Anthropic已否认曾提出担忧或干涉五角大楼使用其技术。

近日双方争端持续升级，政府官员与这家AI公司在社交媒体公开交锋。赫格塞斯本周早些时候曾与阿莫代伊会面，要求该公司在周五前承诺修改合同条款以允许"所有合法用途"。据一位了解会谈情况但无权公开讨论的知情人士透露，赫格塞斯在会上赞扬了Anthropic的产品，并表示国防部希望继续与其合作。

部分专家认为，这场争端本质上是理念冲突，而非具体AI部署方式的分歧。前五角大楼新兴技术副助理部长、AI军事应用专家迈克尔·霍洛维茨指出："在我看来这是场不必要的争执，焦点集中在当前并未提上日程的理论性应用场景。"他注意到Anthropic迄今支持国防部提出的所有技术应用方案，并补充道："我的感觉是，双方在当前技术尚未成熟的应用场景上其实存在共识。"

Anthropic创立之初便以"安全为核心构建AI"为理念。今年一月，阿莫代伊在博客文章中论及强大人工智能的风险，特别提到全自主AI控制武器的危险性。"这类武器在捍卫民主方面确有合法用途，"他写道，"但也是危险的双刃剑。"

本文记者帕雷什·戴夫补充报道。
更新于2026年2月27日东部时间晚6:30：本文已补充国防部长皮特·赫格塞斯在X平台发文内容。

英文来源：

US President Donald Trump announced Friday that he was instructing every federal agency to “immediately cease” use of Anthropic’s AI tools. The move comes after Anthropic and top officials clashed for weeks over military applications of artificial intelligence.
"The Leftwing nut jobs at Anthropic have made a DISASTROUS MISTAKE trying to STRONG-ARM the Department of War,” Trump said in a post on Truth Social.
Trump said that there would be a “six month phase out period” for agencies using Anthropic, which could allow time for further negotiations between the government and the AI startup.
The Pentagon and Anthropic did not immediately respond to requests for comment.
Shortly after the President’s announcement, defense secretary Pete Hegseth said that Anthropic would also be designated a “supply chain risk,” a move normally reserved for foreign businesses considered a danger to American national security. The designation will bar the US military and its contractors and suppliers from working with the AI company.
Hegseth also lashed out at Anthropic and its CEO, Dario Amodei, over the company's refusal to agree to its demands. “Cloaked in the sanctimonious rhetoric of ‘effective altruism,' they have attempted to strong-arm the United States military into submission—a cowardly act of corporate virtue-signaling that places Silicon Valley ideology above American lives,” Hegseth wrote on X.
The Department of Defense has sought to change the terms of a deal struck with Anthropic and other companies last July to eliminate restrictions on how AI can be deployed and instead permit “all lawful use” of the technology. Anthropic objected to the change, claiming that it could allow AI to be used to fully control lethal autonomous weapons or to conduct mass surveillance on US citizens.
The Pentagon does not currently use AI in these ways, and has said it has no plans to do so. However, top Trump administration officials have voiced opposition to the idea of a civilian tech company dictating military use of such an important technology.
Anthropic was the first major AI lab to work with the US military, through a $200 million deal signed with the Pentagon last year. It created several custom models known as Claude Gov that have fewer restrictions than its regular ones. Google, OpenAI, and xAI signed similar deals around the same time, but Anthropic is the only AI company currently working with classified systems.
Anthropic’s model is available through platforms provided by Palantir and Amazon’s cloud platform for classified military work. Claude Gov is currently largely used for run-of-the-mill tasks, like writing reports and summarizing documents, but it is also used for intelligence analysis and military planning, according to one source familiar with the situation who spoke to WIRED on condition of anonymity because they are not authorized to discuss the matter publicly.
In recent years, Silicon Valley has gone from largely avoiding defense work to increasingly embracing it and eventually becoming full-blown military contractors. The fight between Anthropic and the Pentagon is now testing the limits of that shift. This week, several hundred workers from OpenAI and Google signed an open letter supporting Anthropic and criticizing their own companies’ decisions to remove restrictions on military use of AI.
In a memo sent to OpenAI staff today, CEO Sam Altman said that the company agreed with Anthropic and also viewed mass surveillance and fully autonomous weapons as a “red line.” Altman added that the company would try to agree to a deal with the Pentagon that would let it continue working with the military, The Wall Street Journal reported.
The public spat between the Pentagon and Anthropic began after Axios reported that US military leaders used Claude to assist in planning its operation to capture Venezuela’s president, Nicolás Maduro. After the operation, an employee at Palantir relayed concerns from an Anthropic staffer to US military leaders about how its models had been used. Anthropic has denied ever raising concerns or interfering with the Pentagon’s use of its technology.
The dispute between Anthropic and the Department of Defense has escalated in recent days, with officials publicly trading barbs with the AI company on social media.
Hegseth met with Anthropic’s CEO, Dario Amodei, earlier this week. He gave the company until Friday to commit to changing the terms of its contract to allow “all lawful use” of its models. Hegseth praised Anthropic’s products during the meeting and said that the Department of Defense wanted to continue working with Anthropic, according to one source familiar with the interaction who was not authorized to discuss it publicly.
Some experts say that the dispute boils down to a clash over vibes rather than concrete disagreements over how artificial intelligence should be deployed. “This is such an unnecessary dispute in my opinion," says Michael Horowitz, an expert on military use of AI and former deputy assistant secretary for emerging technologies at the Pentagon. “It is about theoretical use cases that are not on the table for now.”
Horowitz notes that Anthropic has supported all of the ways the Department of Defense has proposed using its technology thus far. “My sense is that the Pentagon and Anthropic agree at present about the use cases where the technology is not ready for prime time,” he adds.
Anthropic was founded on the idea that AI should be built with safety at its core. In January, Amoedi penned a blog post about the risks of powerful artificial intelligence that touched upon the dangers of fully autonomous AI-controlled weapons.
"These weapons also have legitimate uses in the defense of democracy," Amodei wrote. "But they are a dangerous weapon to wield."
Additional reporting by Paresh Dave.
Update 2/27/26 6:30pm ET: This story has been updated to include additional information from defense secretary Pete Hegseth's post on X.