Anthropic entered a high-stakes confrontation with Washington just as it was preparing for an IPO. According to the source material, the company refused to let Claude be used for mass surveillance of U.S. citizens or for autonomous weapons systems operating without human involvement, and that refusal led to a federal ban and a Pentagon designation as a “supply chain risk”. For a company valued at $380 billion with $14 billion in 2025 revenue, the dispute goes well beyond one government contract.
Two contract limits became the breaking point
The clash traces back to July 2025, when the Pentagon signed contracts worth up to $200 million each with Anthropic, OpenAI, Google, and xAI to integrate advanced AI into U.S. military systems. Anthropic stood apart by adding two written restrictions to its agreement: Claude could not be used for large-scale surveillance of American citizens, and it could not be used in autonomous weapons systems without human intervention.
In February 2026, Defense Secretary Pete Hegseth gave Anthropic CEO Dario Amodei an ultimatum to remove those terms. If the company refused, the Pentagon would cancel the contract and classify Anthropic as a supply chain risk. Amodei said no. In a public statement cited by the source, he said Anthropic understood that military decisions belong to governments, not private companies, but argued that in rare cases AI could damage rather than defend democratic values.
From contract dispute to federal ban
The fallout moved quickly. Pentagon technology official Emil Michael attacked Amodei on social media. The source also says Donald Trump posted on Truth Social calling Anthropic a “radical left woke company,” describing its move as a disastrous mistake, and ordering federal agencies to stop using Anthropic products and complete the transition within six months.
That was followed by the formal supply chain risk label from the Pentagon. The material stresses why that matters: the tag does not only cut off one customer. It creates a compliance problem for any company working with the U.S. military, because those firms now have to show that Anthropic is not embedded in their operations.
The bigger threat is pressure on partners and workflows
On paper, losing a contract worth up to $200 million may look manageable against $14 billion in annual revenue. The source argues that this is the wrong way to measure the damage. The harder hit comes from the ripple effect across Anthropic’s commercial network, especially where Claude is already tied into government-adjacent work.
Amazon Web Services is described as Claude’s main infrastructure base while also serving as the U.S. government’s largest cloud provider. Palantir integrates Claude in work tied to the military and intelligence community. Defense technology company Anduril uses Claude in Pentagon-related projects. Each of those relationships now sits under pressure. Keep Claude and face compliance scrutiny, or replace a model already woven into internal workflows.
An IPO-era gamble on brand and valuation logic
The source frames the episode as a choice between short-term access and long-term narrative control. If Anthropic had backed down, it could have preserved the Pentagon contract and eased the regulatory confrontation. But the company’s central identity over the past three years — a safety-first, responsible AI developer — would have been weakened at the worst possible time.
If Claude were later linked to mass surveillance or autonomous targeting decisions, the damage would not stop at reputation. The source argues that the valuation premium attached to Anthropic’s responsible AI story could also erode. Refusing the demand, by contrast, shuts much of the government market and creates immediate business risk, yet it also turns the company’s safety claims into a very expensive public proof point. Amodei is quoted in the material as saying the company’s valuation and revenue would only rise after taking this stance.
What happens next is still unsettled
The source says Amazon, Palantir, and Anduril had not publicly stated whether they would alter their Claude-related arrangements. It also notes that replacing Claude may not be simple inside the Pentagon itself. Departments using the model reportedly acknowledged that Claude remained one of the strongest available AI systems, while Grok still lagged behind it technically.
There was another twist on the same day. OpenAI CEO Sam Altman publicly said OpenAI and Anthropic shared the same red lines on military AI boundaries. That leaves the government with fewer obvious alternatives if it wants a major AI provider willing to accept every lawful military use without limits.
The source ends on a narrower point than the political fight around it. Anthropic’s move does not slow the wider militarization of AI. What it does change is the company’s own position inside that trend. The cost is already visible; the outcome of the bet is not.