Analysis
A federal court ruled Thursday that the Trump administration can restrict Anthropic's access to certain government contracts after the company declined to enable specific Claude features for defense use, according to The Information. The ruling sides with the Pentagon in a dispute over what capabilities Anthropic is willing to configure into Claude for military and defense-adjacent government use.
Ars Technica's reporting frames the decision more starkly: the court found the administration can effectively limit Anthropic's federal work over its refusal to enable the disputed features, a characterization that goes beyond a narrow contracting technicality into a broader statement about how much latitude AI vendors have to set usage-policy boundaries once the federal government is a counterparty.
A Pattern of Pressure on Anthropic
This is not an isolated incident. Anthropic has drawn sustained attention from the Trump administration over the past several months -- Axios has reported that administration allies are opening a new front against CEO Dario Amodei specifically, casting him as a face of AI "doomerism" in Washington, and the company's public safety-first positioning has repeatedly put it at odds with an administration favoring faster, less-restricted AI deployment across federal agencies.
The timing matters because Anthropic is barely a month removed from an IPO that priced the company near a $2 trillion valuation on the strength of a roughly $65 billion annualized revenue run rate (as of Sep 21, 2026), with federal and enterprise demand central to the pitch bankers made to investors. A court loss on exactly the kind of government relationship that underpinned that pitch is a different order of risk than a lost commercial deal: it sets a precedent other federal agencies can now cite. OpenAI, which is pursuing its own government-adjacent arrangements, and Google DeepMind, with its own public-sector ambitions, will both be watching how much latitude this ruling actually gives agencies to condition contracts on specific model behavior.
However, neither the court filing nor Anthropic's own disclosures break out a Pentagon-specific dollar figure, so the actual revenue at stake is unclear, and it remains an open question whether Anthropic will appeal or instead adjust Claude's configuration for future defense bids rather than lose the business outright. A single ruling is also not the final word -- Anthropic has both the resources and the incentive to keep litigating, and how far the administration can enforce a broader restriction beyond this specific contract dispute is untested.
For now, the practical effect is that Anthropic has a concrete legal loss to show for a safety stance it has marketed as a differentiator -- and every other frontier lab now has a data point on how far the federal government will go to get the AI behavior it wants from its vendors.