路透社
2026年3月11日 美国东部时间晚上8:35 更新于3小时前
节点运行失败
- 第1项，共2项 2020年1月8日，美国弗吉尼亚州阿灵顿五角大楼简报室的讲台后方可以看到五角大楼标志。路透社/Al Drago/文件照片
[1/2] 2020年1月8日，美国弗吉尼亚州阿灵顿五角大楼简报室的讲台后方可以看到五角大楼标志。路透社/Al Drago/文件照片 购买授权，打开新标签页
3月11日（路透社）——今年1月，在美国国防部（Pentagon，简称五角大楼）与AI实验室Anthropic之间爆发了僵局。原因是该AI实验室拒绝放松其系统的安全护栏，五角大楼将其贴上”供应链风险”标签，这使得该公司的政府合同面临风险。
Claude大模型的高管警告称，他们认为这一指定是对其反对在自主武器和国内监视中使用其技术的报复行为，可能导致其2026年营收削减数十亿美元。
路透社伊朗简报通讯会为您提供伊朗战争最新动态和分析。点击此处订阅。
广告 · 滚动继续阅读
法律专家表示，政府援引的法律与Anthropic的行为不匹配，五角大楼内部行为存在矛盾，且有证据表明其决定可能是出于敌意而非安全考虑，这可能削弱政府的立场。
以下是这场持续冲突的时间线：
1月29日：五角大楼与Anthropic就取消安全保障措施发生冲突。取消这些措施可能使政府能够使用其技术自主瞄准武器并进行美国国内监视
2月11日：五角大楼推动包括Anthropic在内的AI公司，在保密环境中提供其AI工具，且不适用该公司对其他用户施加的许多标准限制
2月14日：由于AI实验室坚持对美军使用其模型设置一些限制，五角大楼考虑终止与Anthropic的合作关系
2月23日：美国国防部长彼得·赫格塞斯（Pete Hegseth）召唤Anthropic首席执行官达里奥·阿莫迪（Dario Amodei）前往五角大楼，就Claude在军事方面的使用进行谈判
2月24日：五角大楼要求Anthropic配合，否则将面临包括被标记为供应链风险在内的后果
2月25日：五角大楼要求包括波音公司（BA.N）和洛克希德·马丁公司（LMT.N）在内的国防承包商评估其对Anthropic的依赖程度
2月26日：五角大楼发言人肖恩·帕内尔（Sean Parnell）要求Anthropic允许五角大楼将其技术用于所有合法目的，并给予该公司直到美国东部时间2月27日下午5:01的决定期限
2月26日：Anthropic表示不会同意五角大楼消除其AI系统安全保障措施的要求
2月27日：美国总统唐纳德·特朗普指示所有联邦机构立即停止使用Anthropic的技术
2月27日：赫格塞斯指示美国国防部将Anthropic指定为”对国家安全构成供应链风险”
2月27日：Anthropic表示将在法庭上挑战五角大楼的决定
2月27日：OpenAI宣布达成协议，将技术部署到国防部的保密网络中
2月28日：OpenAI表示，其与五角大楼的最新协议包含三条红线：其技术不得用于大规模国内监视、指挥自主武器系统或任何高风险自动决策
3月2日：美国国务院、财政部和卫生与公众服务部均停止使用Anthropic的Claude
3月3日：洛克希德·马丁公司承诺遵循国防部的指示，法律专家表示这可能预示着国防承包商将从供应链中移除Anthropic工具，以保护其联邦合同
3月4日：美国财政部部长斯科特·贝森特（Scott Bessent）告诉CNBC，该机构将在几天内从政府系统中移除Anthropic
3月4日：大型科技行业组织推动缓和冲突，称供应链风险指定给公司带来不确定性，并可能威胁军方获取最佳产品和服务的渠道
3月5日：美国国防部正式将Anthropic指定为供应链风险
3月6日：亚马逊表示正在帮助客户将国防部工作负载迁移到其云平台上的替代模型，而客户和合作伙伴仍可继续使用Claude处理所有非五角大楼工作负载
3月6日：美国总务管理局制定了民用人工智能合同的严格规则，并终止了Anthropic的OneGov协议——该协议使Claude可供联邦政府使用
3月9日：Anthropic提起诉讼，阻止五角大楼将其列入国家安全黑名单，称该指定违法，并侵犯了其言论自由和正当程序权利
3月9日：Anthropic高管表示，美国政府将该AI公司列入黑名单可能使其2026年营收减少数十亿美元，并造成声誉损害
3月10日：微软公司（MSFT.O）提交支持Anthropic诉讼的简报，称国防部的指定直接影响到它，并需要临时限制令，以避免代价高昂的供应商中断以及对依赖Anthropic的产品进行仓促重建
Anhata Rooprai在班加罗尔报道；Arun Koyyur编辑
我们的标准：汤姆森路透社信托原则，打开新标签页
推荐主题：
世界
How the Anthropic-Pentagon dispute over AI safeguards escalated
By Reuters
March 11, 2026 8:35 PM UTC Updated 3 hours ago
节点运行失败
Item 1 of 2 The Pentagon logo is seen behind the podium in the briefing room at the Pentagon in Arlington, Virginia, U.S., January 8, 2020. REUTERS/Al Drago/File Photo
[1/2]The Pentagon logo is seen behind the podium in the briefing room at the Pentagon in Arlington, Virginia, U.S., January 8, 2020. REUTERS/Al Drago/File Photo Purchase Licensing Rights, opens new tab
March 11 (Reuters) – A standoff erupted between the U.S. Department of Defense and Anthropic in January after the AI lab refused to loosen safety guardrails on its systems, prompting the Pentagon to label it a ‘supply-chain risk,’ and putting the company’s government contracts in jeopardy.
The Claude maker’s executives warned that the designation, which they view as retaliation for opposing the use of their technology in autonomous weapons and domestic surveillance, could slash their 2026 revenue by billions of dollars.
The Reuters Iran Briefing newsletter keeps you informed with the latest developments and analysis of the Iran war. Sign up here.
Advertisement · Scroll to continue
Legal experts suggest the government’s case may be undermined by a mismatch between the law invoked and Anthropic’s conduct, internal contradictions in the Pentagon’s behavior and evidence that its decision may have been driven by animus rather than security.
Here is a timeline of the ongoing conflict:
January 29 The Pentagon and Anthropic clash over eliminating safeguards that could allow the government to use its technology to target weapons autonomously and conduct U.S. domestic surveillance
February 11 The Pentagon pushes AI companies, including Anthropic, to make their AI tools available in classified settings without many of the standard restrictions that the companies apply to other users
February 14 The Pentagon considers ending its ties with Anthropic over the AI lab’s insistence on keeping some limits on how the U.S. military uses its models
February 23 U.S. Defense Secretary Pete Hegseth summons Anthropic CEO Dario Amodei to the Pentagon for talks on the military use of Claude
February 24 The Pentagon asks Anthropic to get on board, or risk consequences, including being labeled a supply-chain risk
February 25 The Pentagon asks defense contractors, including Boeing (BA.N) and Lockheed Martin (LMT.N), to assess their reliance on Anthropic
February 26 Pentagon spokesperson Sean Parnell asks Anthropic to allow the Pentagon to use its technology for all lawful purposes, giving the company until 5:01 p.m. ET on February 27 to decide
February 26 Anthropic says it will not accede to the Pentagon’s request to eliminate safeguards from its AI systems
February 27 U.S. President Donald Trump directs every federal agency to immediately cease all use of Anthropic’s technology
February 27 Hegseth directs the U.S. DoD to designate Anthropic a “supply-chain risk to national security”
February 27 Anthropic says it will challenge in court the Pentagon’s decision
February 27 OpenAI announces deal to deploy technology in the DoD’s classified network
February 28 OpenAI says its latest agreement with the Pentagon includes three red lines: its technology cannot be used for mass domestic surveillance, to direct autonomous weapons systems or for any high-stakes automated decisions
March 2 The U.S. Departments of State, Treasury and Health and Human Services move to cease using Anthropic’s Claude
March 3 Lockheed Martin pledges to follow the DoD’s direction, signaling a likely exodus of defense contractors removing Anthropic’s tools from their supply chains to protect their federal contracts, legal experts say
March 4 U.S. Treasury Secretary Scott Bessent tells CNBC that the agency will remove Anthropic from its government systems within days
March 4 Big tech industry group pushes to de-escalate the clash, saying a supply-chain risk designation creates uncertainty for companies and could threaten the military’s access to the best products and services
March 5 The U.S. DoD formally designates Anthropic as a supply-chain risk
March 6 Amazon says it is helping customers transition DoD workloads to alternative models on its cloud, while customers and partners could continue using Claude for all non‑Pentagon workloads
March 6 The U.S. General Services Administration draws up strict rules for civilian artificial-intelligence contracts and terminates Anthropic’s OneGov deal, which made Claude available to the federal government
March 9 Anthropic sues to block the Pentagon from placing it on a national security blacklist, saying the designation is unlawful and violates its free speech and due process rights
March 9 Anthropic executives say the U.S. government’s blacklisting of the AI firm could cut its 2026 revenue by multiple billions of dollars and cause reputational harm
March 10 Microsoft (MSFT.O)files a brief backing Anthropic’s lawsuit, saying the DoD designation directly affects it and that a temporary restraining order is needed to avoid costly supplier disruptions and rushed rebuilding of products that depend on Anthropic
Reporting by Anhata Rooprai in Bengaluru; Editing by Arun Koyyur
Our Standards: The Thomson Reuters Trust Principles., opens new tab
Suggested Topics:
World
发表回复