In the high-stakes arena of frontier artificial intelligence, few companies have tested the boundaries of government tolerance quite like Anthropic. Its latest run-ins with the Trump administration reveal more than a simple policy disagreement. They expose a fundamental mismatch between a lab built on principles of caution and a political machine that prizes loyalty above all.
The friction didn’t start in June. It traces back to early 2026 when the Pentagon demanded unrestricted access to Anthropic’s models for any lawful military purpose. The company pushed back. It insisted on explicit bans against using its technology for fully autonomous lethal weapons or mass domestic surveillance. A $200 million contract dissolved. Defense Secretary Pete Hegseth labeled Anthropic a supply chain risk to national security, a designation typically reserved for adversaries like Huawei. President Trump directed all federal agencies to stop using the firm’s tools. Federal News Network reported the move escalated an unusually public clash over AI safety.
Short. Brutal. Costly.
Yet the drama reignited this month. On June 12 the Commerce Department ordered Anthropic to block all foreign nationals from accessing its newest systems, Mythos 5 and Fable 5. The models, one tightly controlled for vetted partners and the other a safeguarded public variant, had to be disabled entirely to comply. Concerns centered on a reported jailbreak in Fable that might expose Mythos’s advanced cyber capabilities, alongside worries about potential access by entities linked to China. Anthropic complied. Then it dispatched senior technical staff, including cybersecurity researcher Nicholas Carlini, to Washington for urgent talks.
By June 26 relief arrived, but only partial. The administration loosened restrictions on Mythos 5, permitting access for a select group of more than 100 trusted companies and agencies. Fable 5 stayed locked down. Commerce Secretary Howard Lutnick wrote in a letter to Anthropic’s chief compute officer Tom Brown that “since the issuance of my June 12 letter, Anthropic has worked with the U.S. government to address risks associated with the Covered Models. These efforts have yielded significant progress.” The Politico account detailed how the partial rescission followed tense negotiations and a shared framework for assessing jailbreak risks.
Anthropic welcomed the step. “We received notice from the US government that Mythos 5, our strongest cybersecurity model, can be redeployed to a small group of cyber defenders and infrastructure providers,” the company said. “We are pleased to see this progress and continue to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.” OpenAI, by contrast, voluntarily withheld wide release of its GPT-5.6 model at the government’s request the same week, prompting its own blog post warning that such interventions “should not become the long-term default” because they keep powerful tools from users who need them.
But here’s the sharper truth. This latest episode fits a pattern of missteps that have left Anthropic politically exposed. CEO Dario Amodei once described Trump as a “feudal warlord” in a deleted post urging votes for Kamala Harris. His sister and co-founder Daniela Amodei contributed to Harris’s campaign. The firm hired former Biden administration officials including those who shaped export controls the current White House later criticized. It declined to donate heavily to Trump-aligned causes. And it held firm on its “red lines” for model use even as rivals showed more flexibility.
Contrast that approach with the standard playbook other tech giants adopted. Meta’s Mark Zuckerberg appointed Trump allies to key roles, donated to the inauguration, and adjusted content policies. Amazon and Apple leaders made their own overtures. OpenAI’s leadership cultivated ties through donations and hires respected in libertarian tech circles. Those moves, however grating in private, bought breathing room. Anthropic chose differently. It bet that technical integrity and public statements about AI risks would suffice. In Trump’s Washington, that calculation has proven expensive.
The Pentagon dispute carried real consequences. Hegseth called Amodei “an ideological lunatic” in congressional testimony. Emil Michael, undersecretary of defense for research and engineering, posted that Amodei “is a liar and has a God-complex.” Trump himself lashed out on social media, accusing the company of trying to “strong-arm the Department of War.” The Fortune analysis laid bare how these attacks stand apart. No other major AI player has faced such sustained vitriol from administration figures.
And the company paid in court as well as public perception. Anthropic sued to overturn the supply-chain-risk designation. A federal judge in San Francisco granted a preliminary injunction in March, finding the government’s actions appeared punitive rather than security-driven. The D.C. Circuit later paused that relief. Litigation continues. Meanwhile the firm’s valuation hovers near $965 billion with an IPO on the horizon. Continued tension could complicate investor appetite.
Yet signs of pragmatic accommodation have emerged. In late June California Governor Gavin Newsom struck a deal to deploy Claude across state agencies and local governments, potentially serving as a template for other states seeking efficiency gains. The New York Times noted the partial loosening of Mythos restrictions de-escalates the immediate feud even if broader questions about government oversight linger. That report highlighted how the administration is carving out a more active role in vetting advanced model releases across the industry.
Amodei has defended the stance. In interviews he argued that crossing certain lines contradicts American values. The firm maintains its constitutional principles for Claude limit harmful applications while still allowing broad utility. Supporters in the AI research community praise the refusal to engage in what one leaked internal message called “safety theater.” Critics inside the administration see arrogance and an agenda to shape regulation in ways that favor incumbents.
So what does this saga signal for the future of AI development in America? The government clearly possesses levers. Export controls, procurement bans, supply-chain designations. Used against a domestic champion, they create uncertainty that ripples through talent markets, customer contracts, and capital allocation. Rivals such as OpenAI appear more attuned to the political currents. They release with restraint when asked. They cultivate relationships. Anthropic’s leaders insist they maintain “lots of friends” in the administration and have joined White House initiatives on energy, education, and youth. Alignment on paper has not translated into smooth operations.
Negotiations continue. Frameworks for evaluating model risks are taking shape. The June 26 letter from Commerce underscores progress on safeguards. But the underlying contest remains. Who ultimately decides when an AI system is ready for deployment? The company that built it, or officials balancing security against innovation speed? Anthropic’s experience suggests that in the current climate, technical arguments alone rarely carry the day. Relationships, signaling, and a willingness to accommodate matter too.
Watch the IPO road map. Watch whether Mythos access expands further or Fable returns. Watch if other labs draw sharper lessons from Anthropic’s bruising encounters. The clash has already reshaped how Washington thinks about its authority over private AI labs. Whether that authority strengthens American leadership or slows it against global competitors will depend on how these tensions resolve in the months ahead. One thing is clear. The old assumption that brilliant models would command automatic deference no longer holds. In this town, politics still shapes technology more than many in Silicon Valley want to admit.
WebProNews is an iEntry Publication