The U.S. government banning the Anthropic model is not really due to any "jailbreaking" reasons.

Author: Zack Whittaker

Translation: Deep Tide TechFlow

Deep Tide Guide: Last Friday, a letter from the U.S. Department of Commerce forced Anthropic to take down its two most powerful models, Fable 5 and Mythos 5. The government’s reason was national security, while outsiders believed it was because the model safeguards were bypassed. But increasing details point to another explanation: this seems more like retaliation after the Trump administration’s relationship with Anthropic soured. A tech company was ordered by the government to shut down products without court approval, sending a signal to the entire U.S. tech industry.

The enforcement letter from the U.S. government to Anthropic, issued before the weekend, forced the company to take all its latest AI models offline. Any U.S. tech company should see this as a warning, not just AI labs.

Let’s review the recent developments. On Friday afternoon, the U.S. Department of Commerce sent a letter to Anthropic, citing a very obscure export control regulation that bans non-U.S. persons (including Anthropic’s own employees) from accessing Fable 5 and Mythos 5, citing unspecified national security concerns. Anthropic said it believes the letter is related to the bypassing of model safeguards, but is not certain because the letter provided no specific details. The letter has not been made public to this day.

Anthropic responded by shutting down both top-tier models for all customers to ensure compliance. The result is that the U.S. government used a swift, unilateral action—seemingly without court approval—to successfully force a tech company to take its models offline.

This intervention by the Trump administration shows that the AI industry cannot be immune to government action. For the broader tech industry, it’s also a warning: either comply or we can shut you and your products down.

Axios quoted sources describing a tense standoff between these two major players over the weekend, saying that the real reason behind the export restriction was a “personality clash” between Anthropic and the Trump administration, rather than any technical issues with the AI products themselves.

New details emerging over the weekend further undermine the already weak justification from the government.

Veteran cybersecurity expert and founder of Luta Security, Katie Moussouris, wrote in a blog that Anthropic recently privately showed her a paper written by several security researchers describing how to bypass the safeguards in Fable 5. (According to The Wall Street Journal, the authors of the paper are security researchers from Amazon.) Moussouris said Anthropic approached her to get her opinion on the paper.

In her blog, Moussouris explained how the researchers triggered the safeguard bypass, but she said this bypass “shouldn’t have triggered the export controls” in the first place. The difference is subtle: asking an AI to “check for security issues in code” versus “fix this piece of code”—the phrasing is slightly different, but the end result is essentially the same.

“The behavior described in the paper can’t really be fixed; any attempt to do so would only weaken the model’s defensive capabilities,” Moussouris said. She criticized the export control order as reckless, crude, and misjudged.

Subsequently, Moussouris and dozens of top security researchers and experts called on the Trump administration to revoke the export restriction, warning that removing advanced cybersecurity capabilities from U.S. cybersecurity defenders is “dangerous.”

Previous administrations have also made sweeping decisions in areas of ignorance. For example, in the 2010s, when the U.S. government revised export laws to regulate cybersecurity tools that could be used for both defense and attack, the wording was too broad, unintentionally risking criminalizing legitimate security and vulnerability research.

But the Trump administration’s order appears to be retaliatory.

Justin Hendrix, editor of Tech Policy Press, said that this move “likely signals to foreign governments that the reliability of U.S. AI in critical scenarios is in question.” The message being sent is: U.S. AI companies cannot operate without interference from the U.S. government.

The Trump administration has not confirmed why it used this export control order. Did officials misread the report and panic? Did Amazon CEO Andy Jassy, out of caution or personal grievance, say something to senior officials that triggered this response? Was there a translation error, or was this a tactic to pressure Anthropic—especially given the already tense relationship? It’s also possible that the White House didn’t realize the letter’s demands would cause such a chain reaction, and officials are now scrambling to clean up the mess they created.

As Hendrix put it, “The current atmosphere is one of suspicion, with senior officials seemingly choosing targets based on personal and political factors.” The consequence is that the government has set a dangerous precedent for how much control it plans to exert over software developed in the U.S.

This time, the target was Anthropic, but tomorrow it could be any other company.