更新于：2026年2月24日 / 美国东部时间下午4:15 / CBS新闻
五角大楼与Anthropic公司之间围绕AI模型使用的信任正在破裂，知情人士向CBS新闻透露。
消息人士称，周二上午在五角大楼的会议中，国防部长彼得·赫格塞斯向Anthropic首席执行官达里奥·阿莫代伊下达了命令，要求其在本周末结束前提交一份签署文件，允许军方全面访问该公司的人工智能模型。
他们表示，官员们正在考虑援引《国防生产法》，以迫使Anthropic遵守军方的要求。Axios此前曾报道过会议中的部分情况。
消息人士告诉CBS新闻，国防官员希望全面控制Anthropic的AI技术，以用于军事行动。该公司于7月获得五角大楼2亿美元合同，用于开发提升美国国家安全的AI能力。
消息人士称，Anthropic多次要求国防部同意设置护栏，限制名为Claude的AI模型对美国人进行大规模监控。国防官员指出这是非法行为，并表示军方只是要求获得严格用于合法活动的使用许可。
自由新闻：[我们是否正处于AI的悬崖边缘？]
一位知情人士称，阿莫代伊还希望确保五角大楼不会在没有人类参与的情况下，将Claude用于军事行动中的最终目标决策。该人士表示，Claude并非没有幻觉风险，其可靠性不足以避免潜在致命错误，如未经人类判断导致的意外升级或任务失败。
但当被要求置评时，一位五角大楼高级官员表示：“这与大规模监控和使用自主武器无关。五角大楼只下达合法命令。”
该官员称，由埃隆·马斯克旗下xAI公司拥有的Grok系统已同意在机密环境中使用，其他AI公司也在接近达成协议。
在周二的会议中，赫格塞斯告诉阿莫代伊，当政府购买波音飞机时，航空航天公司无权干涉五角大楼如何使用这些飞机。他认为军方使用Claude也应遵循同样原则。
阿莫代伊离开后，官员们讨论了在这种情况下是否使用《国防生产法》，该法允许政府对国内产业施加控制。
但两位消息人士称，由于官员们表示目前无法信任Anthropic，五角大楼可能会正式将该公司列为“供应链风险”，以迫使它退出政府项目。Anthropic是首家获授权参与军方机密网络工作的科技公司。
Anthropic发言人在声明中表示：“我们继续就使用政策进行真诚对话，以确保Anthropic能够根据我们模型可靠且负责任的能力，继续支持政府的国家安全任务。”
赫格塞斯给了Anthropic周五下午5点的最后期限。
Hegseth demands full military access to Anthropic’s AI model Claude and sets deadline for end of week
Updated on: February 24, 2026 / 4:15 PM EST / CBS News
Trust is breaking down between the Pentagon and Anthropic over the use of its AI model, sources familiar with the situation told CBS News.
In a meeting at the Pentagon Tuesday morning, Defense Secretary Pete Hegseth gave Anthropic’s CEO Dario Amodei until the end of this week to give the military a signed document that would grant full access to its artificial intelligence model, the sources said.
Officials are considering invoking the Defense Production Act to make Anthropic adhere to what the military is seeking, they said. Axios reported earlier on some of what transpired in the meeting.
Defense officials want full control of Anthropic’s AI technology for use in its military operations, sources told CBS News. The company was awarded a $200 million contract by the Pentagon in July to develop AI capabilities that would advance U.S. national security.
Anthropic has repeatedly asked the Defense Department to agree to guardrails that would restrict the AI model, called Claude, from conducting mass surveillance of Americans, sources said. Defense officials noted that that’s illegal and said the military is simply asking for a license to use the AI strictly for lawful activities.
*
The Free Press:[Are We at an AI Precipice?]
*
Amodei also wants to ensure Claude is not used by the Pentagon for final targeting decisions in military operations without any human involvement, one source familiar with the meeting said. Claude is not immune from hallucinations and not reliable enough to avoid potentially lethal mistakes, like unintended escalation or mission failure without human judgment, the person said.
But when asked for comment, a senior Pentagon official said: “This has nothing to do with mass surveillance and autonomous weapons being used. The Pentagon has only given out lawful orders.”
The official said Grok, which is owned by Elon Musk’s xAI, is on board with being used in a classified setting, and other AI companies are close.
In Tuesday’s meeting, Hegseth told Amodei that when the government purchases Boeing planes, the aerospace company has no say in how the Pentagon uses the planes. He argued the same should be true for the military’s use of Claude.
After Amodei left, officials discussed whether to use the Defense Production Act in this situation, which enables the government to exert control over domestic industries.
But because officials say they aren’t sure the government can trust Anthropic at this point, the Pentagon may decide to officially designate the company as a “supply chain risk” to push them out of government, two sources said. Anthropic was the first tech company authorized to work on the military’s classified networks.
An Anthropic spokesperson said in a statement, “We continued good-faith conversations about our usage policy to ensure Anthropic can continue to support the government’s national security mission in line with what our models can reliably and responsibly do.”
Hegseth gave Anthropic a deadline of 5 p.m. Friday.
发表回复