Anthropic built its reputation around AI safety. Now its work with the Pentagon is becoming a test of how far that promise can go when military needs, classified systems, and lethal operations enter the same conversation.
A classified AI partnership becomes a pressure point
Last year, Anthropic became the first major AI company cleared by the US government for classified use, including military applications. The move did not create a large public shock at the time, but a new dispute has made the relationship far more consequential.
The Pentagon is now reconsidering its relationship with Anthropic, including a $200 million contract. The reported reason is that the company objects to taking part in certain deadly operations.
The so-called Department of War may also designate Anthropic as a “supply chain risk.” In the context described by the source article, that label is usually associated with companies that do business with countries scrutinized by federal agencies, like China. If applied here, it would mean the Pentagon would not work with firms that use Anthropic’s AI in defense projects.
Chief Pentagon spokesperson Sean Parnell confirmed to WIRED that Anthropic was under pressure. “Our nation requires that our partners be willing to help our warfighters win in any fight. Ultimately, this is about our troops and the safety of the American people,” he said.
Why Anthropic’s stance matters
The case matters because Anthropic is not just another AI vendor. It has positioned itself as the most safety-conscious of the major AI labs, with guardrails meant to prevent harmful uses of its models.
The company provides the government a “custom set of Claude Gov models built exclusively for U.S. national security customers.” At the same time, Anthropic has said it is doing so without violating its own safety standards, including a ban on using Claude to produce or design weapons.
Anthropic CEO Dario Amodei has specifically said he does not want Claude involved in autonomous weapons or AI government surveillance. That position now appears difficult to maintain inside a defense environment that wants fewer restrictions on how AI can be used.
The dispute also sits alongside another reported issue: whether Anthropic is being punished for complaining that Claude was used as part of the raid to remove Venezuela's president Nicolás Maduro. The company denies that account. The source also notes that Anthropic publicly supports AI regulation, a stance described as unusual in the industry and at odds with the administration’s policies.
The Pentagon is sending a wider message
This is not only about Anthropic. OpenAI, xAI and Google currently have Department of Defense contracts for unclassified work and are pursuing the clearances needed for classified use.
For those companies, the Anthropic dispute may define the terms of future defense work. The Pentagon’s message is direct: AI partners must be prepared to support the military’s ability to win.
Department of Defense CTO Emil Michael told reporters that the government would not tolerate an AI company limiting how the military uses AI in its weapons. He raised the example of a drone swarm and asked what options exist if human reaction time is not fast enough.
That question captures the central conflict. AI companies often describe guardrails as essential to preventing harm. Military officials may view those same guardrails as constraints that reduce effectiveness in combat.
AI safety meets military reality
The source article frames the dilemma through a broader idea: many researchers and executives believe AI may be the most powerful technology ever invented. Current AI companies were largely founded around the belief that AGI, or superintelligence, could be developed while preventing widespread harm.
That belief is hard to square with systems built for military and intelligence operations. Isaac Asimov’s first law of robotics is quoted in the source: “A robot may not injure a human being or, through inaction, allow a human being to come to harm.”
The tension is especially sharp when AI leaders also believe their systems may become smarter than any human on Earth. If guardrails are supposed to remain reliable under that future, the decision to build versions of AI for national security customers becomes more than a business strategy. It becomes a test of whether safety commitments can survive pressure from the state.
There is also a practical national security argument. The source notes that the best technology from innovative companies can be valuable to defense. It also observes that tech companies that once hesitated to work with the Pentagon are, in 2026, generally eager to become military contractors.
Palantir CEO Alex Karp is cited as being direct about this reality, saying, “Our product is used on occasion to kill people.” The source contrasts that openness with the lack of similar public clarity from AI executives about models associated with lethal force.
The risk of an AI arms race
The larger implication is that national security AI may not remain a one-sided advantage. The source argues that sophisticated opponents will need to build their own versions, pushing the world toward a full-tilt arms race.
In that setting, governments may have little patience for companies that draw fine distinctions around legal use or lethal practices. The Pentagon’s position, as described in the source, is that companies wanting defense partnerships must be ready to do what it takes to win.
That mindset may fit military logic, but it creates a serious problem for AI safety. A company trying to build AI that does not harm people faces an obvious contradiction if it also develops versions connected to lethal force.
The source notes that only a few years ago, governments and tech executives were discussing international bodies that might monitor and limit harmful AI use. That conversation is now much quieter. The future of warfare is increasingly treated as AI-driven, and the future of AI itself may be shaped by the demands of warfare unless companies and nations contain the technology carefully.