Anthropic's refusal to bend to Washington has cost it Pentagon contracts and earned it a court fight it did not expect

The Trump administration designated Anthropic a national security supply-chain risk, banned federal agencies from using its products, and forced it to pull its newest AI model after the company refused to allow unrestricted military use of its Claude models. The unprecedented actions have cost Anthropic Pentagon contracts and sparked a legal battle, while rivals OpenAI and Google secured federal deals by cooperating with Washington.

The Trump administration designated Anthropic a national security supply-chain risk, forced it to pull its newest AI model, and banned federal agencies from using its products. No other American company has ever received that label. On February 27, 2026, President Trump directed every federal agency to stop using Anthropic's products. Defense Secretary Pete Hegseth went further, formally designating the company a "supply-chain risk to national security," a label the U.S. government had previously reserved for organizations tied to foreign adversaries, most notably Huawei. Anthropic, a San Francisco AI lab founded on the premise that powerful models require rigorous safety constraints, had refused Hegseth's deadline to allow unrestricted military use of its Claude models, specifically rejecting autonomous weapons applications and mass domestic surveillance. That refusal triggered the most aggressive government action ever taken against a domestic AI company. The commercial fallout was immediate. Defense contractors who use Anthropic's models in their Pentagon work are now required to certify they don't, or risk losing their contracts. A federal judge called the designation "troubling" in March and granted a preliminary injunction blocking its enforcement, but the U.S. Court of Appeals for the D.C. Circuit denied Anthropic's request to stay the supply-chain label while the litigation continues. As of late June, the dispute remains unresolved in court. Then, in mid-June, a second confrontation landed. As Fortune reported, Amazon CEO Andy Jassy alerted White House officials that researchers had found a way to bypass guardrails in Fable 5, Anthropic's newest model. Within 90 minutes, according to a source familiar with the company, the Commerce Department invoked export controls to pull Fable 5 and its underlying Mythos 5 model from distribution, just three days after launch. It was the first time the U.S. government has used export controls to halt a commercial AI model already in public hands. Anthropic pushed back, arguing that a narrow potential jailbreak does not meet any coherent threshold for invoking national security powers, and that applying this standard industry-wide would effectively freeze all frontier model releases. NPR reported the episode left considerable confusion inside Washington about what the administration's actual AI regulation framework is, because there isn't one. While Anthropic was fighting subpoenas and export orders, OpenAI and Google were attending White House AI summits, softening safety rhetoric, and securing federal procurement relationships. OpenAI's GPT-5.6 and Google's models entered Pentagon and intelligence community pipelines. The contrast is not subtle. The companies that played the access game got contracts. The company that didn't got designated a supply-chain threat alongside Huawei. Dario Amodei has been explicit about his position. In a June 10 blog post titled "Policy on the AI Exponential," he called for mandatory third-party testing of frontier models across four risk categories: cybersecurity, biological weapons, loss of AI control, and automated R&D that could accelerate the other three. That puts Anthropic directly at odds with an administration whose instinct, as Council on Foreign Relations analysts noted, is to sustain a deregulatory, innovation-first posture while reaching for security controls whenever a specific threat surfaces. The result is incoherence: the same White House that stripped AI oversight in January 2025 is now shutting down AI models by executive fiat sixteen months later. For founders building on Claude and enterprise buyers who have bet on it, this is a material risk question, not a philosophical one. Anthropic has told investors roughly 80% of its revenue comes from enterprise customers, a base Amodei described as relatively stable. That stability is now being tested. The supply-chain designation requires defense-adjacent vendors to certify they don't use Claude. Any enterprise with significant government exposure has a real compliance problem on its hands, regardless of what a district court judge thinks. The partial thaw is real but thin. Trump told CNBC in late June that Anthropic is "shaping up" after Amodei met with Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent to discuss Mythos, the company's AI cybersecurity tool. Trump has said a Pentagon deal is "possible." But the supply-chain designation is still in effect, the export restriction on Fable 5 is still in effect, and the appeals court has not moved. A presidential comment to cable news is not a policy reversal. The harder question for investors pricing Anthropic's $61 billion valuation is whether principled positioning is a long-term brand asset or a near-term commercial ceiling. The honest answer is probably both, and the ratio depends on who wins the next election. Enterprise buyers in financial services, healthcare, and legal, sectors with no Pentagon exposure, have actually moved toward Claude since February. The public response was striking: Claude became the most downloaded free app on the U.S. Apple chart on February 28, the day after the federal ban. Anthropic's refusal to fold played well with the part of the market that does not need a security clearance to buy software. But the government market is foreclosed for now, and the precedent the administration has set is worrying regardless of which lab you're building on. TechPolicy.Press noted that the Mythos recall reveals an administration trying to balance a deregulatory posture with novel security concerns by improvising in real time, which is a different and more dangerous thing than having a policy. If Anthropic wins its court case and a future administration normalizes relations, the brand bet pays off. If the export controls expand, the litigation drags, and the Pentagon contracts stay locked, the $61 billion number will need a harder look. Also read: Jane Street and TSMC's venture arm have backed Etched, the startup betting that locking silicon to transformers beats Nvidia on inference economics https://startupfortune.com/jane-street-and-tsmcs-venture-arm-have-backed-etched-the-startup-betting-that-locking-silicon-to-transformers-beats-nvidia-on-inference-economics/ • Central banks are warning that AI leverage is the real risk, not just the valuations https://startupfortune.com/central-banks-are-warning-that-ai-leverage-is-the-real-risk-not-just-the-valuations/ • Apptronik is turning robot training into the next AI factory https://startupfortune.com/apptronik-is-turning-robot-training-into-the-next-ai-factory/