Anthropic said no. The Pentagon demanded full access to Claude for military operations. The company refused to budge on two red lines: domestic mass surveillance and fully autonomous weapons. President Trump called them a “Radical Left AI company” and banned all federal agencies from using their products. Defense Secretary Pete Hegseth designated them a “supply chain risk” - a designation previously reserved for foreign adversaries.
Then, hours after the ban, the military used Claude in combat operations against Iran anyway.
This contradiction - banning a company while simultaneously relying on its technology for active military operations - exposes the chaos and confusion at the heart of AI governance in 2026.
What Anthropic Actually Refused
According to CBS News, Anthropic pushed for explicit contractual safeguards preventing two specific uses: mass surveillance of American citizens and autonomous weapons systems without human oversight.
CEO Dario Amodei told CBS that Anthropic agreed to “98%-99% of the military’s use cases.” The sticking point wasn’t military use broadly - it was these two narrow categories.
“We are patriotic Americans,” Amodei told Fortune. “Everything we have done has been for the sake of this country, for the sake of supporting U.S. national security.”
On mass surveillance, Amodei’s concern centers on AI capabilities outpacing legal frameworks. Surveillance that was technically infeasible - and therefore didn’t need explicit prohibition - is now possible. He argues this “domestic mass surveillance is getting ahead of the law.”
On autonomous weapons, his position is more pragmatic than principled: current AI systems aren’t reliable enough. He cited “basic unpredictability” in AI models as a technical limitation requiring human oversight. Anthropic isn’t categorically opposed to autonomous weapons, but believes the technology isn’t ready.
The Pentagon’s position was simpler: existing laws already prohibit these activities, so explicit contractual restrictions were unnecessary. They wanted access for “all lawful purposes.”
How Claude Was Used in Iran
Despite the ban, the military continued using Claude during weekend strikes on Iran. According to CBS News, two sources confirmed Claude supported operations including “synthesizing documents and making logistics and supply chains more efficient.”
Pentagon CTO Emil Michael defended the continued use: “At some level, you have to trust your military to do the right thing.”
The timing is striking. Trump announced the ban on Friday. By Saturday, Claude was being used in active combat operations. Either the ban was never intended to apply to existing military systems, or different parts of the government simply aren’t coordinating.
The implications for AI-enabled warfare are significant. Craig Jones, a warfare expert, told Fortune that AI has dramatically compressed military decision timelines. The targeting process that once took months during Vietnam can now happen in near real-time.
“The AI machine is making recommendations for what to target, which is actually much quicker in some ways than the speed of thought,” Jones said.
OpenAI’s Different Path
While Anthropic faced blacklisting, OpenAI struck a deal with the Pentagon - announced hours after Anthropic’s designation as a supply chain risk.
OpenAI agreed that its technology would not be used for “domestic mass surveillance” or “autonomous weapon systems,” affirming that humans would take “responsibility for the use of force.” On paper, these restrictions mirror what Anthropic wanted.
But the deal faced immediate backlash. CEO Sam Altman admitted the deal was “definitely rushed” and conceded that “the optics don’t look good” and that it “looked opportunistic and sloppy.”
OpenAI subsequently amended the contract to clarify that “the AI system shall not be intentionally used for domestic surveillance of U.S. persons and nationals.”
The difference between the two companies’ outcomes - blacklisting versus contract - may come down to negotiating posture rather than substance. Anthropic demanded explicit guarantees before signing. OpenAI signed first and added clarifications after public pressure.
The Worker Rebellion
The most significant development may be the employee response. Tech workers at Google and OpenAI organized around an open letter titled “We Will Not Be Divided.”
According to CNBC, the letter grew from a couple hundred signatures on Friday to nearly 900 by Monday - close to 800 from Google and nearly 100 from OpenAI.
“They’re trying to divide each company with fear that the other will give in,” the letter reads. “That strategy only works if none of us know where the others stand.”
This echoes 2018, when over 4,000 Google employees signed an internal petition opposing Project Maven, a Pentagon contract for drone targeting technology. About a dozen employees resigned in protest. Google ultimately chose not to renew the contract.
Whether current protests achieve similar results depends on scale. The 900 signatories are a fraction of Google’s 180,000+ workforce. But the movement is cross-company in a way that 2018’s wasn’t, potentially harder for individual companies to dismiss.
The Consumer Response
In an unexpected twist, Anthropic’s principled stance has been excellent marketing. Claude surged to #1 on the App Store over the weekend, dethroning ChatGPT.
An Anthropic spokesperson told Axios that daily sign-ups have broken records every day since the confrontation began. Free users increased over 60% since January. Paid subscribers more than doubled. The surge even caused temporary outages from “unprecedented” demand.
The consumer market is voting with their downloads - and they’re rewarding the company that said no to the government.
What the Rules Actually Say
The Pentagon’s policy on autonomous weapons, DOD Directive 3000.09, updated in January 2023, requires “appropriate levels of human judgment over the use of force.” It distinguishes between:
- Human in the loop: Semi-autonomous systems where operators select targets
- Human on the loop: Supervised systems where operators can intervene
- Human out of the loop: Fully autonomous systems that select and engage targets independently
The directive requires human oversight and the ability to intervene, but the precise meaning of “appropriate levels” remains contested. And importantly, these are Pentagon policies, not laws - they can be changed by future administrations.
Anthropic’s concern is that even with current policy requiring human oversight, the combination of political pressure and technical capability could erode these safeguards over time.
The Bigger Picture
This confrontation reveals a fundamental tension in AI governance. As The Conversation notes, “ethical military AI assumes it is operating under democratic principles.”
Anthropic’s insistence on negotiating ethical boundaries represents “democratic practice in action.” The administration’s punitive response - blacklisting a company for disagreeing - signals intolerance for that practice.
The irony of using Claude in combat hours after banning its maker suggests the government needs these AI tools regardless of political posturing. That leverage could, in theory, allow companies to maintain ethical boundaries. In practice, it may just mean the government finds more compliant partners.
The Bottom Line
The Anthropic standoff isn’t really about one company or one contract. It’s about whether AI developers can set meaningful limits on how their technology is used - and whether governments will respect those limits or simply find partners who won’t ask questions.
The next few months will determine whether this becomes a turning point or a footnote. If tech worker protests spread and consumer preference continues rewarding ethical stances, companies may find principles profitable. If the government successfully isolates Anthropic while rewarding OpenAI’s accommodation, the message will be clear: compliance over conviction.
For now, Anthropic remains blacklisted. Claude remains in military service. And the contradictions remain unresolved.