Anthropic’s Washington rift deepens as AI controls ease

The Commerce Department has eased export restrictions on Anthropic’s Mythos AI model after a two-week block that forced the company to disable it for users, Fortune reported. The reversal came the same day OpenAI said it would hold back a broad release of GPT-5.6 at the U.S. government’s request, showing how differently two leading AI labs are being treated in Washington.

Fortune reported that Anthropic’s conflict with the Trump administration has become unusually sharp for a major technology company. The company, valued at $965 billion and preparing for an expected IPO in the coming months, has faced two government moves that Fortune described as potentially severe threats to its business.

In April, the Pentagon designated Anthropic a supply chain risk after the company would not accept contract terms the department wanted, according to Fortune. Two weeks before the Commerce decision, the government imposed export controls on Mythos and Fable, a related model meant for broader commercial use, after a Fable jailbreak raised concerns that users could bypass safeguards and reach Mythos’s stronger cyber capabilities.

Administration criticism has been unusually personal

Trump administration officials have repeatedly attacked Anthropic and CEO Dario Amodei, Fortune reported. President Donald Trump wrote on Truth Social that the company was made up of left-wing extremists trying to pressure the Pentagon, which Trump has renamed the Department of War.

Fortune also cited posts and remarks from senior officials. Emil Michael, undersecretary of defense for research and engineering, accused Amodei on X of lying and having a God complex, while Defense Secretary Pete Hegseth called Amodei an ideological lunatic during an April congressional hearing. David Sacks, Trump’s former AI and crypto czar who remains involved in government technology advisory committees, has accused Anthropic of using AI safety concerns to advance regulation and of trying to promote woke AI rules.

During the export-control dispute, unnamed senior U.S. officials portrayed Amodei as slow to respond to the White House, Fortune reported. Anthropic disputed that account, saying Amodei spoke with the administration within one hour and 15 minutes of the call.

A different posture from other tech companies

Fortune framed the dispute around Anthropic’s refusal to use the same approach as several other technology companies seeking favor with Trump’s team. Meta CEO Mark Zuckerberg, Amazon founder Jeff Bezos and Apple CEO Tim Cook have made gestures or personnel moves viewed as friendly to the administration, according to Fortune.

OpenAI has also developed ties in Trump’s political circle, Fortune reported. The publication said OpenAI policy chief Chris Lehane has standing among some Trump allies from his work for technology and crypto interests, and OpenAI cofounder Greg Brockman has become the largest donor to the pro-Trump super PAC MAGA Inc.

Anthropic took a different path. Fortune reported that Amodei once described Trump in a deleted Facebook post as a feudal warlord while urging friends to vote for Kamala Harris, and that cofounder Daniela Amodei donated to Harris and previously worked for Hillary Clinton.

The company also hired several former Biden administration AI officials, including Ben Buchanan, Elizabeth Kelly and Tarun Chhabra, Fortune reported. It later withdrew legal work from Skadden and Latham & Watkins after those firms reached settlements with Trump, with Amodei reportedly telling staff he would not work with firms he believed had yielded to attacks on the rule of law.

Safety stance wins support but carries risk

Anthropic has not cut off the White House, Fortune reported. Amodei attended a Trump energy event in Pennsylvania in 2025, traveled with Trump to Japan, backed the White House AI Action Plan, joined an AI education event and signed the administration’s Pledge to America’s Youth.

Still, Fortune reported that Amodei was left out of a September White House dinner for major tech and AI leaders and was not among executives invited on Trump’s state visit to the U.K. that month. Amodei told Fortune in October that Anthropic would speak up when it disagreed with the White House, even if agreement might help the business.

The company’s refusal to accept Pentagon contract language without explicit bans on use of its models for autonomous weapons or domestic mass surveillance drew praise from many AI researchers, including at rival labs, Fortune reported. Fortune also reported that some consumers moved from OpenAI’s ChatGPT to Anthropic’s Claude after OpenAI accepted Pentagon terms allowing use of its models for any lawful purpose.

Anthropic has since hired Ballard Partners and added Republicans and former Trump officials to its policy operation, Fortune reported. Brendan Steinhauser, CEO of the Alliance for Secure AI, told Fortune that Anthropic may have underestimated the importance of relationships and money in Washington, while saying the company’s stance fits its culture and Amodei’s public identity.

This story draws on original reporting from Fortune.