Anthropic employees accuse the Trump administration of targeting them

Executives at artificial intelligence startup Anthropic received alarming news from the White House on Friday. They were told they had less than 90 minutes to take down their latest AI models due to national security concerns.

Inside the company, private employee group chats immediately lit up. Managers were instructed to prepare customers for potential service disruptions to the models, called Fable 5 and Mythos 5. But the messaging kept changing, with workers initially told the security issue was the ability of foreign companies to gain access to the systems, and later that a major vulnerability had been discovered in the models.

In employee chats, Anthropic engineers asked each other whether the company’s plan to go public this year would run afoul of White House guidelines. Many shared reports that offered conflicting information about why the White House ordered Anthropic to suspend access to Fable 5 and Mythos 5 for all foreign nationals.

“What do you tell your clients?” asked one staffer in a chat room following The New York Times. Another said: “Does anyone know what to believe?” In another message, one worker said, “I don’t understand what the problem is.”

Six days later, Anthropic’s roughly 3,000 employees still have few answers. The San Francisco company continues to grapple with internal turmoil as Dario Amodei, the CEO, and some of his representatives meet with the Trump administration to try to resolve the situation. But after discussions on Monday and Tuesday, there was no breakthrough on ending a US order restricting access to the company’s new AI models.

In a statement Monday, Anthropic said it would continue to meet with government officials and pledged its “continued commitment to work with the administration.”

The dispute highlights how unique Anthropic has become in Washington. It was the second time in six months that a fast-growing artificial intelligence startup has been embroiled in a battle with the Trump administration over its powerful technology, even as other AI companies offer similar models that have not received the same attention. And it left Anthropic employees in what they called a holding pattern, with some wondering if President Trump had hand-picked them.

“Are we being bullied based on bad vibes?” one employee asked in a chat shown by The Times.

A White House official familiar with the discussions said the administration continues to work with Anthropic to address its concerns. Anthropic’s Mr. Amodei will attend the Group of 7 summit in Évian-les-Bains, France this week.

Antropic and the Trump administration first clashed this year over a $200 million Defense Department contract for artificial intelligence in classified systems. In a highly public argument, the company and the Pentagon disagreed on how the technology should be used in war.

This led Defense Secretary Pete Hegseth to label the Anthropocene a “supply chain risk” in February. The designation, which meant the company posed a national security risk, had never been used against a US company. Anthropic sued the government over the label.

The company’s relationship with the Trump administration changed in April when it announced Mythos, a new model of artificial intelligence. Mythos was so strong at identifying security vulnerabilities in software, Anthropic said, that it could trigger a “cybersecurity showdown.” The start-up would retain the model, he added, except for a select few organizations and companies.

At the White House, it prompted officials to begin discussing an executive order on AI that would include a voluntary process for companies to review new AI models before release. Antropic has been involved in discussions surrounding the order and spoke with White House chief of staff Susie Wiles in a meeting that some hoped would lead to a rapprochement.

Then this month, Anthropic released a straitjacketed version of the Mythos, Fable 5, which researchers believed was safe for widespread use and included additional handrails. Antropic provided the models to the Commerce Department for testing, as it has done with previous models, and was not told not to deploy them, people with knowledge of the matter said. Test results are unknown.

The release of Fable 5 prompted other companies to try the model. One of them was Amazon, which has pledged to invest up to $33 billion in Anthropic and supplies some of the chips that power its AI models.

Amazon scientists wrote a preliminary document on Fable 5 that pointed to a perceived security flaw, representatives of the Anthropocene Organization, U.S. officials and others said. Using this vulnerability, Amazon was able to convince Fable 5 to expose bugs in specific parts of the vulnerable software code. Amazon did not include tests of models from other AI firms, some of which are capable of generating the same information, said cybersecurity experts who have seen or been briefed on the report.

After Amazon shared the findings with Anthropic, the paper appeared in a previously scheduled call between Amazon and administration officials, according to people briefed on the events. Officials asked Amazon to then call Scott Bessent, the finance minister, to explain the vulnerability in more detail. Amazon agreed.

An Amazon spokesman would not comment on specific questions about the sequence of conversations with the White House, but said in a statement: “It is not unusual for governments to seek our advice on potential security risks. When they arise, we do not share the details of those discussions.”

US officials who reviewed the Amazon document said the results were “terrifying”. However, cybersecurity experts said the ability to ask Fable to identify flaws could be more useful for cyber defense than for malicious hackers. People familiar with the technology said that everything the Amazon paper accused Anthropic of being capable of doing could also be done by OpenAI’s latest model.

Katie Moussouris, a prominent cybersecurity expert who reviewed Amazon’s work at Anthropic’s request, said in a blog post that concerns about Fable need not be addressed because it would undermine its ability to help defenders secure their software code.

“Defenders need to be able to ask AI to fix errors in a file, explain why the fix matters, and write tests to confirm that the fix works,” Ms Moussouris said. “That’s not a guardrail bypass. It’s the most valuable thing an AI model can do for defensive security.”

One government official said the problem with Fable 5 goes beyond Amazon’s paper and includes unspecified national security concerns about which companies Anthropic chose to work with. That was not a point raised directly with Antropic, said three people with knowledge of the discussions.

On Friday, the government called Anthropic and demanded that it pull the latest model off the market and deny access to all clients. Within 15 minutes, the executives were on the phone with U.S. officials asking about national security concerns, three people briefed on the call said.

But no further details were given. The startup learned about Amazon’s research paper on its own, the three people said, adding that Anthropic had reached out to the White House to ask if the paper was cause for concern.

Some cybersecurity experts said the Trump administration is unfairly targeting Anthropic. By Tuesday afternoon, more than 150 of them had signed an open letter to Howard Lutnick, the commerce secretary, and Sean Cairncross, the national cyber director, urging them to lift the restrictions on Anthropic models.

The list of signatories included cybersecurity figures credited with major contributions to Internet security, as well as an AI expert at chip giant Nvidia and a former top National Security Agency official who oversaw the responsible use of AI.

“Anthropic has built several safeguards into the Fable model to prevent it from being used for a cyber offensive,” the letter reads. “These protections were so aggressive that they were a source of humor in the cyber community on launch day.”

At Anthropic, some employees shared the letter among themselves, seeing it as evidence that they were being unfairly targeted. On company channels, workers also asked how their employer plans to move forward if the White House continues to restrict their ability to release new AI models.

“At what point does it feel like they don’t want us to exist?” one employee said in a chat on Tuesday.