‘he Has To Find A Way To Be Friends’: The Political Fight Behind Anthropic’s Export Controls

Aligning your message with the White House is standard political practice for tech companies trying to stay on the right side of fast-changing AI policies in President Donald Trump's Washington.

Anthropic CEO Dario Amodei has not played that game.

Amodei has over the last few months repeatedly diverged from White House messaging in ways that set him apart from other Big Tech CEOs. He clashed with Defense Secretary Pete Hegseth over artificial intelligence’s battlefield role. He warned that the technology could trigger mass job losses as the White House pitched its economic benefits. And, in the latest escalation, he disputed the White House’s account of a security breach on its latest model.

Taken together, it represents an enormous and remarkably public gulf in political sensibilities, between an administration that puts an unusual degree of pressure on big companies to show their political deference, and an independent-minded Silicon Valley phenomenon with relatively little interest in adhering to Washington’s rules.

The result is a strained relationship, such that when Amodei asked Trump officials for time to address what he saw as a small technical problem with his latest model, the administration refused.

Instead, the administration instituted export controls on that model last Friday, prompting Anthropic to take down Fable 5, the company’s latest and most powerful product — a drastic move that reverberated through the industry and left each side baffled by the other.

Anthropic has in many ways worked with the White House. Amodei attended an energy event with the president in Pittsburgh last summer to support the administration’s vision of American energy and AI dominance. The company supported the White House’s AI Action Plan, a move the administration highlighted as evidence of its rapturous reception across the industry. It signed on to the White House’s Pledge to America’s Youth, and participated in the White House's AI Education Taskforce event. Amodei joined Trump in Japan last fall and has made regular trips to Washington to meet with members of Congress and the administration to discuss national security concerns.

What Anthropic sees in the latest escalation is an administration response that was disproportionate to the severity of the security risk. In the week since, Anthropic staff have met with administration officials in an attempt to address the government’s concerns.

“Both parties are working quickly to get this resolved,” said a spokesperson for the company, who was granted anonymity to describe the negotiations. “This is part of our longstanding commitment to working alongside the administration, including protecting US critical infrastructure and the US lead in cyber defense."

But while the administration is known to have a heavy hand with its enemies, even some critics of Trump’s recent AI policy acknowledge that Amodei did himself and his company no favors.

“This is an example where just keeping your powder dry is immensely useful,” said Dean Ball, a former Trump administration official who has criticized the White House over a number of recent policy decisions and who recently announced a new role at Anthropic’s top competitor OpenAI.

“The worst possible outcome is the thing that's happening right now, where Dario just keeps fighting. I think, unfortunately, the brutal reality is that he has to find a way to be friends, or if not friends, he has to find a way to kick the can down the road,” he said.

In the most recent dispute, critics say it was Amodei’s own words that came back to haunt him. He had repeatedly warned that Anthropic’s most powerful model, Mythos, was too capable of hyper-advanced cyberattacks to be released without stringent controls. Then the company released Fable, a dulled-down version of Mythos, with guardrails that it said would make the model safe for the masses.

“You can’t tell everyone that your product might destroy the world and then not expect the government to be involved,” an administration official said. “They’re politically naive.”

Anthropic worked closely with the administration before it released Fable, it said in a blog post reacting to the export controls, and submitted the model to government agencies for pre-release testing. The company also said it collaborated with the White House as it expanded access to Mythos.

The administration, which is accustomed to political alignment from companies it works alongside, has not been quiet about its displeasure with Anthropic in the aftermath of last Friday’s decision.

“Three months ago, @DeptofWar kicked @AnthropicAI out of our building—forever. Every passing day proves why that was the right move,” wrote Hegseth, who had declared the company a risk to the Pentagon supply chain in March in a dispute over AI’s potential use for mass surveillance and autonomous weaponry.

Former White House deputy chief of staff Taylor Budowich also leaned into the politicization, posting the X profile of one of the cybersecurity experts vouching for Anthropic to show that she’d listed her pronouns and was wearing a hat supporting Democrats in her profile picture.

“These people really just don't get it....,” he said about Anthropic’s decision to request the expert’s help.

Several days after the export controls were imposed, as daily negotiations between the administration and a team of employees from Anthropic continued, Fable remained offline. POLITICO reported Thursday that a resolution could include establishing a technical assessment that both evaluates the newly identified vulnerability that led to the escalation and sets a standard for assessing future security flaws.

Last week’s decision to impose export controls came after tense talks among Treasury Secretary Scott Bessent, National Cyber Director Sean Cairncross and Commerce Secretary Howard Lutnick with Amodei and other senior leaders at the company. Those discussions concerned a breach reported by Amazon, which said it had found a bypass in Fable that allowed people to get around the guardrails Anthropic had set — an issue known in the industry as a jailbreak.

During the calls, Amodei tried to clear up what he assumed was a technical misunderstanding. He defended the model’s guardrails, which Anthropic has said are some of the strictest in the industry, and pushed back on the administration’s concerns, according to a senior White House official who was granted anonymity to discuss the internal deliberations. Officials asked him to immediately take down the model, and when he didn’t agree to do so, Bessent told him that was a “bad decision.”

Then the government imposed the export restriction on Fable, prohibiting Anthropic from allowing non-Americans access to it. The company chose to take it down altogether.

Over the following weekend, a group of technical experts and CEOs wrote an open letter to the administration urging officials to revoke the export control. They said the actions that Amazon researchers had gotten Fable to carry out were a “necessary capability in any model that is intended to write secure code,” and “should not be considered an offensive capability.” 175 tech leaders have now signed it.

Ball noted that both sides underestimated the politics at play, which led to frustrations and ultimately a breakdown in communication.

“You can't ignore the political dimension of this conflict,” he said. “There are surely other dimensions, like the national security dimensions, but you can't ignore the political dimension of this conflict, and I think there are missteps that both sides in this have made.”

Ryan Fedasiuk, a tech fellow at the conservative America Enterprise Institute, went a step further — accusing the administration of leaning headfast into politics and abandoning its policy.

“Until Friday, this administration has been unapologetically pro-AI, pro-innovation, and pro-diffusion,” Fedasiuk wrote on X. “It's not worth sacrificing that position to politically assassinate Anthropic.”

Some within the AI industry fear that the simmering dispute, which has lasted more than a week, will goad the administration into a broader approach of heavy-handed regulation

“Not enough people being wise, patient, and thoughtful in this moment. Too many people putting on war paint and seeing issues du jour as the important battle lines. No. The stage is way bigger,” OpenAI’s chief futurist Joshua Achiam wrote on X.

Achiam said he was concerned that the disagreement over Fable and the subsequent export restriction was “normalizing” the dramatic step of “electronic citizenship verification as a step in using software,” setting a dangerous precedent.

The two companies have a storied rivalry: Amodei and others formed Anthropic after he left OpenAI over disagreements with CEO Sam Altman. And OpenAI has broadly taken a much different approach to dealing with the Trump administration.

After Trump’s 2024 victory, Altman distanced himself from his past support for Democrats, and the company was fast to hire former Trump campaign and administration officials. OpenAI also struck a deal early in Trump’s term to fund a massive buildout of data centers, called Stargate, a move in line with the president’s push for a rapid expansion of AI.

Meanwhile, a super PAC funded by OpenAI co-founder Greg Brockman has funneled money to candidates opposing AI regulation, aligning with the Trump administration’s politics. (A super PAC funded by Anthropic sprung up in defiance, instead backing candidates who support AI safety policies).

The competing companies are both headed toward blockbuster IPOs, with market values that each could land in the neighborhood of $1 trillion.

In a Wednesday interview with Bloomberg, Amodei shed light on his past disagreements with Altman that led him to leave the company.

"We'll see who wins in the market and we'll see who wins in the court of public opinion," Amodei said. "I think those things speak louder than any drama about why who left what."

In the past, Amodei’s approach has won over AI skeptics and people who want to see the industry consider safety above speed. The dispute with the Pentagon landed Anthropic’s Claude as the top-ranking App store download, bumping OpenAI’s ChatGPT and Google’s Gemini.

But the disparate approaches have led to markedly different relationships with the president: Altman sat next to Trump at this week’s AI lunch during the G7 in France. Amodei, who also attended, sat next to France’s President Emmanuel Macron.

An OpenAI official, granted anonymity to speak about the dynamics, said OpenAI has “really tried to spend time with the administration, from the top on through, in educating folks on where we see the technology going.”

“We try to over communicate in terms of what we're doing and how we're doing it and what we're seeing,” the official said. “As a result of that, we have established I think a really strong professional relationship, and that's not just to the administration, we do this both on the Democratic and Republican side on the Hill.”

Anthropic, too, has enlisted scores of lobbyists and public affairs specialists in Washington, including Trump-linked lobbying group Ballard Partners, and has tripled its presence both last year and this year as it seeks a better relationship with the administration, said a person close to Anthropic who was granted anonymity to describe the company’s efforts.

“As DC grapples with the policy impacts of AI, Anthropic is making our biggest institutional bet yet here: tripling our policy team, opening a permanent DC office, and making a bipartisan, long-term investment in our Washington AI policy operation,” the person said.

The company has hired former Trump administration officials, including Chris Liddell, a former deputy chief of staff in Trump’s first White House and Mary Croghan, who served in the Office of Cabinet Affairs, also in Trump’s first term.

Anthony Cimino, Anthropic’s head of federal affairs, served on the Republican House Committee on Financial Services, and a number of other Republicans, including former Missouri Sen. Roy Blunt, serve on Anthropic’s national security advisory board.

But it’s the CEO, Amodei, who ultimately delivers the company’s message.

Brendan Steinhauser, a longtime conservative who runs a lobbying group pushing for AI regulation, the Alliance for Secure AI, said Anthropic’s history with the administration comes down to a risk calculus Amodei has decided he’s comfortable with.

“Anthropic’s just kind of said this is what we want to do, these are our values and principles, and they've been very open about it,” Steinhauser said.

“Whereas the other companies, for the most part, have said, OK, we're going to play the game, and we're going to say yes, and do whatever [the administration] wants to do, because if we don't, we're going to be paying the consequences.”