Why Anthropic Is Fighting With Trump (Again)

The artificial intelligence firm announced late Friday that it was disabling its two newest large language models just three days after launching them, citing new export controls that had just been imposed by the U.S. government that prevent any foreign nationals from using those models.

The abrupt export control decision has exacerbated ongoing tensions between the Silicon Valley firm and the Trump administration over the national security concerns around the use of Anthropic’s AI models, and it sets up a second high-profile showdown between them in less than a year.

Anthropic executives reportedly held talks with administration officials in Washington on Monday to try to find a resolution, but no agreement has yet been reached. Neither side responded to requests for comment on what was discussed.

Here’s what you need to know.

What happened?

Anthropic received an export control directive from the U.S. government at 5:21 p.m. EDT on Friday that cited “national security authorities” to bar any foreign nationals (including non-U.S. Anthropic employees) from using the company’s Claude Fable 5 and Claude Mythos 5 models, the company said in a statement. It added that the directive “did not provide specific details of its national security concern.”

Anthropic said the government had verbally expressed concerns about vulnerabilities in the models that could allow users to “jailbreak” or bypass their safety guardrails. However, the company claimed that those vulnerabilities “appear[ed] relatively simple” and could also be uncovered by other publicly available models, such as OpenAI’s ChatGPT.

The White House did not immediately respond to Foreign Policy’s request for comment on its specific concerns, which were reportedly flagged to the administration by Amazon. Amazon shared its concerns with Anthropic before expressing them to the U.S. government, according to a person familiar with the matter.

The tech giant is a major investor and partner of Anthropic as well as a key provider of cloud computing services to the U.S. government. “As a leading cloud provider that serves a large number of private and public sector customers, it’s not uncommon for governments to seek our counsel on potential security risks,” an Amazon spokesperson said in a statement. “When they occur, we don’t share the details of these discussions.”

Rather than trying to hive off access just for U.S. users, Anthropic opted to shut down the models altogether. “The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance,” the company said, noting that access to the company’s other models would remain unimpeded.

Wait, why does “Mythos” sound familiar? 

Anthropic raised alarm bells back in April when it unveiled an earlier version of one of its new models, known as Claude Mythos Preview. At the time, the company warned that the model was too powerful for a full public release because it had shown a preternatural ability to detect cybersecurity flaws in computer systems—thus theoretically making it easier for hackers to infiltrate them. Instead, Anthropic gave a few dozen (and growing) companies and government agencies access to Mythos so they could use it in a closed environment and patch their systems before the broader release.

That broader release came last Tuesday, when Anthropic announced Claude Fable 5—a model it said was built off Mythos but with additional safeguards and restrictions on certain queries that made it “safe for general use.” It also announced Claude Mythos 5, a slightly tweaked version of the model without some of those safeguards that could only be used by “a small group of cyberdefenders and infrastructure providers.”

The big rollout was short-lived, however, with the Trump administration putting the kibosh on both models just three days later.

This isn’t the first time the Trump administration has tangled with Anthropic, is it? 

Nope. Back in February, Anthropic got into a massive fight with the U.S. Department of Defense over the U.S. military’s deployment of the company’s models. Anthropic demanded that the Pentagon commit to barring those models from being used for autonomous weapons or domestic surveillance.

U.S. Defense Secretary Pete Hegseth responded by giving Anthropic an ultimatum: Give the military unfettered access to Anthropic’s technology or be designated a “supply chain risk”—a label that would prevent Anthropic from serving any customers who work with the Pentagon.

Hegseth subsequently made good on that threat, prompting the company to sue the U.S. government in response. That lawsuit is still ongoing, even as Anthropic and the Trump administration fight a new fight this week.

“Three months ago, [the Defense Department] kicked @AnthropicAI out of our building—forever,” Hegseth wrote in a post on X on Saturday. “Every passing day proves why that was the right move.”

What does this mean for the rest of the world? 

The Anthropic-Trump fight has placed further strain on an already rocky trans-Atlantic relationship, with European countries vying for carve-outs that would allow them to keep using the most advanced U.S. AI models.

European diplomats reportedly held talks with U.S. Commerce Secretary Howard Lutnick on the sidelines of the G-7 summit in Évian-les-Bains, France, this week, seeking to negotiate “trusted partner” access, the Financial Times reported. The United Kingdom has also requested special dispensation, though a Trump administration official told the New York Post that exempting any country from the export controls, including allies, would be “completely illogical.”

Canadian Prime Minister Mark Carney warned that the U.S. export controls show the risks of global dependence on U.S. technologies. “The situation we’re in collectively right now with Mythos and Fable is something that can happen with overreliance on certain models,” Carney said during a visit to Ireland ahead of the G-7 summit. “Nobody has done anything wrong in the situation,” he added. “But we will have done something wrong if we just accept this, don’t take the lesson, don’t build out and diversify.”