In contract negotiations between senior Defense Department officials and leaders from AI giant Anthropic in December, the company agreed to allow the U.S. government to use its AI systems for missile and cyber defense purposes, a person familiar with the matter said, requesting anonymity to speak about private discussions.
But that apparently did not satisfy the Pentagon.
Following weeks of tension between the Defense Department and Anthropic over the company’s restrictions on how its products can be used by the military, Defense Secretary Pete Hegseth issued a stark ultimatum to company CEO Dario Amodei on Tuesday: Allow the AI technology to be used for all legal military purposes by this Friday or be forced to cooperate, a senior Pentagon official told NBC News.
The ultimatum, detailed to NBC News by a senior Pentagon official, comes as Anthropic — a company that has heavily marketed its focus on AI safety — tries to maintain firm policies preventing its systems from being used for mass domestic surveillance or direct use in lethal autonomous weapons.
The December contract changes would allow for its systems to be widely used for cyber and missile defense, according to the person familiar with the matter. An Anthropic spokesperson told NBC News in a statement that “Every iteration of our proposed contract language would enable our models to support missile defense and similar uses.”
But the company’s insistence on guardrails have continued to be a source of contention between Anthropic and the Defense Department.
According to the senior Pentagon official, representatives from the department, including Undersecretary of Defense Emil Michael, recently discussed several hypothetical scenarios with Anthropic leadership about how the company’s products might be employed by the military.
As part of those discussions, the officials discussed how Anthropic’s systems might be used if an adversary launched an intercontinental ballistic missile at the U.S. According to the Pentagon source, the officials discussed whether Anthropic’s guardrails might somehow block a U.S. response to the launch. Anthropic officials said they could be called on to lift those restrictions, according to the official, but Pentagon leadership was not fully satisfied with Anthropic’s adjustments and did not want to be beholden to the private company.
According to an Anthropic spokesperson, any suggestion that CEO Amodei said the Pentagon would have to call the company in each missile defense operation is “patently false.”
In the latest escalation in negotiations, during Tuesday’s meeting Pentagon leaders said they could invoke the Defense Production Act to force Anthropic to comply with the Pentagon’s rules, according to the senior Pentagon official. The Act allows the president to control domestic companies critical to national security in times of need.
In Tuesday’s meeting, Pentagon leadership also invoked threats to instead label Anthropic as a “supply chain risk” and ban all defense business with the company if it does not align its terms of service for certain high-stakes uses with the Pentagon by Friday, the source said.
“Anthropic has until 5:01pm Friday to get on board with the Department of War,” the senior Pentagon official said of the ultimatum in a statement provided to NBC News, responding to questions about the meeting. “If they don’t get on board, the Secretary of War will ensure the Defense Production Act is invoked on Anthropic, compelling them to be used by the Pentagon.”
“Additionally, the Secretary of War will also label Anthropic a supply chain risk,” the official said.
Asked about Tuesday’s meeting, an Anthropic spokesperson said in a statement: “Dario expressed appreciation for the Department’s work and thanked the Secretary for his service. 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 complimented Anthropic’s products and said the Pentagon wanted to work with Anthropic, according to another person familiar with the meeting, who requested anonymity to speak candidly. The person confirmed that the department said it would terminate Anthropic’s work with the Pentagon by Friday if it did not agree to its terms.
According to reports from The Wall Street Journal and Axios, Anthropic’s Claude systems were used during the operation to capture Venezuelan President Nicolás Maduro in January. It is unclear exactly how the systems were used.
Hegseth sent a memo to senior Pentagon officers Jan. 9 announcing the Pentagon’s drive toward an “AI-first warfighting force.” He outlined a push to use AI models, like Anthropic’s, for all legitimate military purposes, “free from usage policy constraints” set by individual AI companies.
Anthropic is the only AI company whose products are actively used on classified networks, through its contract with Palantir, a data analytics company. A senior Pentagon official confirmed to NBC News that xAI reached a deal with the Pentagon on Monday to use its Grok chatbot system on classified networks, agreeing to allow its systems to be harnessed for “any lawful use” as Hegseth desired.
Anthropic was one of four AI companies — the others were OpenAI, Google DeepMind and xAI — to get contracts worth up to $200 million in July to “prototype frontier AI capabilities that advance U.S. national security.”
About the Author
