Trump administration officials began weighing sanctions on Anthropic weeks before they demanded the company take its latest and most advanced artificial intelligence model offline, after a dispute shattered the White House’s already-fragile trust in the company, according to two White House officials who spoke on the condition of anonymity to describe private deliberations.
Several weeks ago, Anthropic gave the administration a list of 111 organizations slated to receive advanced access to Mythos, its newest model, one of the officials said. With companies around the world clamoring for the technology after Anthropic warned it could dramatically enhance hackers’ capabilities, officials reviewed the list and signed off.
But Anthropic later disclosed that the list had ballooned and roughly 50 additional entities had already received access. Senior officials began to consider using export controls to claw back the technology after the company did not identify the new recipients for days.
When Anthropic finally turned over the names, the administration discovered that one recipient was a South Korean telecommunications company the administration suspected of having ties to China, the officials said.
Anthropic quickly revoked the firm’s access to Mythos, said a person close to Anthropic, who spoke on the condition of anonymity to describe the deliberations.
But the episode badly damaged officials’ confidence in the company’s ability to safeguard sensitive technology.
“They expanded it too far and wide,” one of the officials said.
Officials at the Defense Department, CIA and National Security Agency pushed to move ahead with export controls, but others in the administration cautioned against taking such a drastic step, one official said.
So when Amazon alerted the administration last week to a security vulnerability in Anthropic’s new model, officials saw it as the latest in a series of lapses. They concluded it was time to ask President Donald Trump to authorize export controls, forcing Anthropic to pull the model just three days after it became widely available to customers.
The person close to Anthropic said that the Trump administration threatened export controls for the first time on Friday in the wake of the Amazon findings. The company did not withhold the name of a partner with Mythos access and shared it with national security partners, the person added.
(Amazon Executive Chairman Jeff Bezos owns The Washington Post.)
The showdown marks a striking escalation in the Trump administration’s approach to artificial intelligence, after previously pushing a light-touch regulatory approach. By using national security authorities to force a leading U.S. AI company to withdraw a flagship product, the White House has signaled a willingness to intervene directly in the development and distribution of cutting-edge AI models, potentially setting a new precedent for the government’s oversight of the industry’s most powerful systems.
The dispute has also brought together some administration officials who have spent months battling over the government’s approach to Mythos, uniting them against a common enemy: Anthropic, one of the officials said.
The White House declined to comment.
Commerce Secretary Howard Lutnick, who was largely sidelined from the Mythos regulation debate, is also now involved because his department issues export controls, the officials said. He is now presenting himself as Anthropic’s friend advocating to save an innovative American company, one official said.
The administration wants to lift the sanctions as soon as it’s convinced Anthropic has fully accounted for everyone who received the Mythos product and has strengthened the protections on the Fable project to prevent it from being jailbroken, the official said. The administration has no intention of moving against other products from Anthropic or other AI companies, the person added.
“They dug their own grave,” the official said, claiming the company had not complied with a voluntary testing regime that had been established through a recent executive order.
On Friday morning the administration asked Anthropic to send technical experts to Washington and they didn’t arrive until Sunday night, concerning officials that the company wasn’t treating the situation as an emergency, one of the officials said.
Senior technical staff met with administration officials virtually throughout the weekend before their trip to Washington, according to an Anthropic spokesperson.
Cybersecurity experts, tech executives and some venture capitalists expressed alarm at the government’s intervention, calling on the administration to reverse course in a letter publishedSunday. They warned that security research should be used to improve products, not to ban them.
“There was a speeding ticket, and they gave Fable the death penalty,” Alex Stamos, the former chief security officer of Facebook, said in an interview. Stamos was listed as the first signatory of the letter.
In the meantime, experts say, the administration’s approach risks undermining the nation’s technological leadership if businesses around the world begin to wonder whether building on top of American AI systems could leave them vulnerable to being left hanging at a moment’s notice if the administration suddenly issues export controls.
Michael Semick, a tech entrepreneur in Seattle, has been using Fable to help build the software for his new marketing technology startup. The model is expensive to use, but he found it far outperformed Anthropic’s previous models and ones from competing tech companies at certain longer reasoning tasks.
But now that the tech has been pulled by Anthropic, he’s had to make do with alternatives.
“It’s not going to stop my business but what it will do is slow me down,” Semick said. For the government “to come in and place a restriction on use that basically renders it unusable is disruptive to the entire ecosystem of innovation.”
Tensions have been growing between Anthropic and some of Trump’s top AI advisers for years. The company has positioned itself as a champion for regulation, and it was supportive of some of the measures the Biden administration took to rein in the AI industry. Several former Biden officials now work for Anthropic.
The company’s approach leaves it philosophically opposed to David Sacks and some of Trump’s other allies in Silicon Valley, who have advocated for an acceleration of AI development. These tensions exploded into public view earlier this year when the company foughtwith the Pentagon over how its technology could be used.
Sacks, the administration’s former AI and crypto czar, wrote on X that the administration issued the export control “reluctantly.”
“The Admin values Anthropic’s technical capabilities and feels that this issue, while serious, should be easily resolved,” he wrote. “The ball is in Anthropic’s court.”
The post How Anthropic lost the White House’s trust — and then its flagship product appeared first on Washington Post.
