The United States military has formally opened its most sensitive digital corridors to Elon Musk’s artificial intelligence company xAI, signing an agreement to deploy the Grok model inside classified systems where the nation’s most consequential work gets done — intelligence analysis, weapons development, and the real-time management of battlefield operations.
The deal, confirmed by a Defense Department official to Axios and first reported by The New York Times, arrives at a moment of maximum pressure on xAI’s chief rival in the classified space: Anthropic, whose flagship model Claude has until now been the only AI permitted inside those systems, and whose CEO was summoned to the Pentagon on Tuesday for what senior officials described, with unmistakable clarity, as anything but a routine conversation.
The timing is not a coincidence. The Pentagon did not simply decide to diversify its classified AI portfolio out of good strategic hygiene. It decided to do so because Anthropic has refused to give the military what it wants — unconditional access to Claude for any lawful purpose, including applications the company’s own ethics framework explicitly prohibits.
The two uses Anthropic has drawn firm lines against are mass surveillance of American citizens and the development of fully autonomous weapons capable of firing without a human in the decision loop. The Pentagon’s position, in the words of its chief spokesman Sean Parnell, is straightforward: “Our nation requires that our partners be willing to help our warfighters win in any fight.” xAI, by contrast, has agreed to the military’s “all lawful use” standard without conditions. That agreement is now signed.
The standoff escalated dramatically in the weeks following January 3, when U.S. special operations forces conducted a raid in Venezuela that resulted in the capture of President Nicolás Maduro. Claude was accessed through Anthropic’s partnership with Palantir during the operation, making Anthropic the first known AI developer whose technology was used in a classified operation by the U.S. Department of War, though it remains unclear how the tool was deployed.
What happened next is where the accounts diverge. A senior Pentagon official told Fox News Digital that an Anthropic executive raised a question with a Palantir executive about whether Claude had been used in the raid — a question phrased in a way that implied Anthropic might disapprove if it had been. The Palantir executive, alarmed, informed the Pentagon. Anthropic disputes that characterization, describing the outreach as part of a routine technical discussion. The Pentagon did not accept that explanation.
A senior Pentagon official did not mince the institutional response: “It will be an enormous pain in the ass to disentangle, and we are going to make sure they pay a price for forcing our hand like this.”
The weapon the Pentagon has been preparing is a “supply chain risk” designation — a label that is typically reserved for foreign adversaries and would carry consequences far beyond the cancellation of Anthropic’s existing contract. Such a designation would require the plethora of companies that do business with the Pentagon to certify that they don’t use Claude in their own workflows — a sweeping demand given that Anthropic recently said eight of the ten biggest U.S. companies use Claude. The contract at stake is valued at up to $200 million, a relatively small figure for a company with $14 billion in annual revenue, but the secondary effects of a supply chain blacklisting could be far more damaging.
Against that backdrop, Defense Secretary Pete Hegseth scheduled Tuesday’s meeting with Anthropic CEO Dario Amodei. A senior Defense official framed the session in terms that left no room for diplomatic ambiguity: “Anthropic knows this is not a get-to-know-you meeting. This is not a friendly meeting. This is a sh*t-or-get-off-the-pot meeting.”
Hegseth was prepared to deliver an ultimatum: accept the military’s terms or face the full range of consequences the Pentagon has been assembling. The question before Amodei was not just whether to keep a contract. It was whether the government can compel a private AI developer to disable its own safety architecture as the price of doing business with the military.
The deeper culture clash here is worth examining without the lazy framing that legacy media tends to apply. Amodei and Anthropic are not pacifists sabotaging national security out of naïve idealism. The company has been unambiguous about its commitment to U.S. national security — Claude was, after all, the first commercial AI model cleared for classified military networks, a distinction Anthropic earned deliberately. A source familiar with the negotiations described Anthropic’s position as not ideological but focused on two particular issues: that AI systems are not reliable enough for life-or-death decisions without human judgment, and that the technology significantly changes what is possible with domestic surveillance.
Those are serious technical and constitutional arguments, not Berkeley faculty lounge politics. The problem is that the Pentagon, operating under an administration that has little patience for Silicon Valley ethics seminars, views the same arguments as corporate overreach into sovereign military decision-making.
Meanwhile, xAI’s entry into the classified space reshapes the competitive landscape for AI in national security. Google is reportedly close to a deal allowing classified use of its Gemini model, while OpenAI remains not close as it continues working on safety technology. The race into classified systems is accelerating precisely because the prize is enormous — not just contract dollars, but the opportunity to define how artificial intelligence operates at the most consequential levels of American power.
xAI’s willingness to accept the “all lawful use” standard without conditions makes it the most accommodating partner the Pentagon currently has among frontier AI developers. Whether Grok’s capabilities can fully match Claude’s in the specialized context of classified defense applications is a separate and unresolved question. Senior administration officials acknowledged that competing models “are just behind” when it comes to specialized government applications, complicating any abrupt switch.
There is an irony embedded in this story that deserves acknowledgment. Anthropic was founded specifically because its founders believed that building powerful AI without adequate safety constraints was an existential risk to humanity. That mission — taken seriously, not as a marketing slogan — has led the company to hard limits on autonomous weapons and mass surveillance. Those limits now put it on a collision course with the U.S. government, facing a designation that treats it with the same suspicion as a foreign adversary.
Whatever one thinks of the Pentagon’s demands, the spectacle of an American AI safety company being threatened with national security blacklisting for insisting that its technology not be used to surveil Americans without human oversight is not a story that fits comfortably into any simple narrative. The meeting between Hegseth and Amodei on Tuesday will not fully resolve these tensions — it will only determine whether the argument continues inside the relationship or outside of it. In either case, the question of who controls the rules of AI in classified military systems has moved from theory to immediate, high-stakes reality.
