The U.S. government’s executive letter to Anthropic just before the weekend, effectively forcing the company to take its latest AI models offline, should be a wake-up call to all U.S. tech companies, including the AI Institute.
Read about the news barrage: On Friday afternoon, the U.S. Department of Commerce sent a letter to Anthropic requesting vague export control guidelines that would ban non-Americans, including Anthropic’s employees, from accessing Fable 5 and Mythos 5, citing unspecified national security concerns. Anthropic said it believes the letter is related to the model’s guardrail circumvention, but is not sure because the letter does not provide specific details. The letter has not been made public.
In response, Anthropic shut down both top models for all customers to ensure compliance with the guidelines. As a result, the U.S. government successfully forced tech companies to take their models offline through rapid, unilateral actions that did not appear to require court approval.
The Trump administration’s intervention on Friday shows that the AI industry is not immune from government interference. It’s also a warning to the wider tech industry. Failure to comply could result in closure of you and your product.
Citing sources, Axios described the tense situation between the two major companies over the weekend, saying “personality differences” between Antropic and the Trump administration, rather than technical issues with the AI product, led to the export order.
New details about the matter that emerged over the weekend have now cast further doubt on the government’s already shaky reasoning.
Katie Moussouris, a cybersecurity veteran and researcher who founded Luta Security, said in a blog post that Anthropic recently shared with her a personal copy of a paper written by security researchers describing alleged guardrail bypasses in Fable 5. (The Wall Street Journal reports that the paper’s author is a security researcher at Amazon.) Moussouris said Anthropic contacted him to request his opinion on the paper.
Moussouris’ blog post explained how the researchers triggered the guardrail bypass, but said the bypass itself “should not have triggered the export controls.” The difference is between asking an AI model to “review your code for security issues” and asking it to “fix this code.” Even though the questions are posed slightly differently, the end result is much the same.
Mussouris criticized the export control guidelines as hasty, coercive and wrong, saying, “The actions described in the document cannot be meaningfully corrected and any attempt will only undermine the national defense model.”
Moussouris and dozens of other top security researchers and experts later called on the Trump administration to rescind the export control order, calling the move to withdraw advanced cybersecurity capabilities from America’s network defenders “dangerous.”
Past administrations have made sweeping decisions about knowledge gaps. For example, language used by the U.S. government in the 2010s to amend export laws regarding cybersecurity tools that could also be used in cyberattacks was so broad that it inadvertently almost made legitimate security and vulnerability research illegal.
But the Trump administration’s directive appears to be retaliatory.
Justin Hendricks, editor-in-chief of Tech Policy Press, said the Trump administration’s actions “are likely to raise alarms in foreign capital about the reliability of U.S. AI for critical applications.” The message is that American AI companies cannot be trusted to operate without U.S. government interference.
The Trump administration did not provide a reason for issuing the export control directive. Did government officials misread the report and get scared? What did Amazon CEO Andy Jassy say to senior government officials that prompted a response out of caution or malice? Was something lost in translation? Or was it a way for the administration to put pressure on Anthropic, with which it already had a fraught relationship? The White House failed to recognize the far-reaching consequences of the letter’s demands, and officials are likely scrambling to undo damage of their own making.
In Hendrix’s words, “there is a cloud of suspicion that senior officials are choosing their favorites based on personal and political factors.” The fallout is that it sets a dangerous precedent for how much control the government will exert over the release of American-made software.
This time the government took issue with Anthropic. Tomorrow I might be with someone else.
If you purchase through links in our articles, we may receive a small commission. This does not affect our editorial independence.
