Author: Zack Whittaker

Translated by: Shenchao TechFlow

Shenchao Guide: Last Friday, a letter from the U.S. Department of Commerce forced Anthropic to take down its two strongest models, Fable 5 and Mythos 5. The government cited national security as the reason, while outsiders suspected it was because the model safeguards were bypassed. However, more and more details point to another explanation: this seems more like a retaliation following a rupture in the relationship between the Trump administration and Anthropic. The fact that a tech company could be forced to shut down products by a government order without court approval sends a signal to the entire U.S. tech industry.

The enforcement letter sent to Anthropic by the U.S. government, a prelude to the weekend, compelled the company to take down all its latest AI models. Any U.S. tech company should treat this situation as a warning bell, not just AI labs.

First, let's sort through the news of the past few days. On Friday afternoon, the U.S. Department of Commerce sent a letter to Anthropic, citing a rare export control regulation that prohibits non-Americans (including Anthropic's own employees) from accessing Fable 5 and Mythos 5, due to some unspecified national security concerns. Anthropic stated that it believed this letter was related to the model safeguards being bypassed, but was also uncertain, as the letter did not provide specific details. This letter has not been made public to date.

In response, Anthropic took the step of taking both top models down for all customers completely to ensure compliance. The result was that the U.S. government managed to compel a tech company to take its models offline with a swift, unilateral action that apparently did not require court approval.

This intervention by the Trump administration indicates that the AI industry cannot be immune to government scrutiny. This is also a warning for the broader tech industry: comply, or we can shut you and your products down.

Axios cited sources describing a tense situation between these two major players over the weekend, stating that the real reason for the export order was the "personal mismatch" 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 flimsy rationale provided by the government.

Cybersecurity veteran, researcher, and founder of Luta Security, Katie Moussouris, wrote in a blog that Anthropic recently showed her a paper written by several security researchers, which described the so-called bypass of the safeguards in Fable 5. (According to a report by the Wall Street Journal, the paper's author is a security researcher at Amazon.) Moussouris indicated that Anthropic approached her to get her opinion on the paper.

Moussouris explained in the blog how the researchers triggered this safeguard bypass but stated that the bypass itself "should not have triggered export controls." The difference is minimal: asking the AI to "check for security issues in the code" versus asking it to "fix that segment of code" is slightly different in phrasing, but the end result is essentially the same.

"The behavior described in the paper cannot truly be fixed; any attempt to do so would only weaken the model's defensive capabilities," Moussouris said. She criticized the export control order as rash, blunt, and misjudged.

Subsequently, Moussouris, along with dozens of top security researchers and experts, called on the Trump administration to revoke this export control order, claiming that withdrawing advanced cybersecurity capabilities from U.S. cyber defenders is "dangerous."

Previous administrations have made blanket decisions based on knowledge gaps. For instance, when revising export laws in the 2010s, the U.S. government aimed to regulate cybersecurity tools that could be used for both defense and offense, but the language was too broad, almost categorizing legitimate security and vulnerability research as unlawful.

However, this order from the Trump administration appears to be retaliation.

Justin Hendrix, editor at Tech Policy Press, noted that the Trump administration's action "is likely to alarm foreign governments about the reliability of U.S. AI for critical scenarios." The message conveyed is that U.S. AI companies cannot operate without interference from the U.S. government.

The Trump administration has not confirmed why it invoked this export control order. Was it a case of officials misreading that report and panicking? Did Amazon CEO Andy Jassy, out of caution or personal grudge, inform government officials about something, sparking this reaction? Was there a mistranslation, or was it inherently a pressure tactic against Anthropic—after all, relations had already been tense? It's also possible that the White House simply did not realize that the demands of this letter would trigger such a significant chain reaction, and officials are now scrambling to fix the situation they inadvertently created.

In Hendrix's words, "the current atmosphere is one of suspicion, with senior officials seemingly selecting favored targets based on personal and political factors." The consequence is that the government has set a dangerous precedent regarding the level of control it intends to exert over U.S.-made software.

This time, the government is targeting Anthropic, but tomorrow it could be anyone else.