Amazon’s warning to Trump officials helped spark a US crackdown on Anthropic’s juiciest AI toys.
According to the Wall Street Journal, the Trump administration’s decision to halt all foreign use of Anthropic’s most capable AI models followed chats between Amazon chief executive Andy Jassy and US officials, including US Treasury secretary Scott Bessent, people familiar with the matter said.
Amazon researchers had used a series of prompts to get Anthropic’s Fable 5 model to cough up information that could help cyberattacks, Jassy told officials. That information was supposed to be off-limits.
Tech industry executives have been nattering regularly with the administration about the power of cutting-edge AI tools.
Soon after Jassy’s warning, White House officials met to discuss a response, while security researchers began testing Amazon’s claims. Officials asked Anthropic to fix the vulnerabilities or take down the model, administration officials said.
They decided the bluntest fix was to stop foreign governments, companies and individuals from accessing the tool. President Donald Trump later approved the move, despite worries that it could put a dent in innovation.
The administration had long thought Anthropic, one of the leaders in America’s AI race, could not be trusted to manage the security risks around its new model. A Friday call between officials and Anthropic chief executive Dario Amodei apparently did little to soothe nerves.
Some officials said Anthropic seemed unwilling to work with government security experts to address the problem.
Anthropic said it was cutting off access to Mythos and Fable for all users to ensure compliance. That could knacker companies worldwide using the tools to find software vulnerabilities.
Part of the problem was that the US government was insisting that only true red, white, and blue Americans could work on the tools, and Anthropic said many of its researchers are foreign-born, meaning the rules effectively stopped them from working on the latest models.
Anthropic has said the vulnerabilities flagged by Amazon are relatively basic. The company said other publicly available models can find them and that they do not amount to a full jailbreak.
Some security researchers familiar with Amazon’s work agree with that view.
The startup insists it has adequate safeguards and is known for prioritising safety. It previously delayed wider access to Mythos at White House direction and works with a key government AI testing unit before releasing models.
A senior White House official said the intervention was about model safety and did not undermine the administration’s innovation push. White House AI adviser and venture capitalist David Sacks, an Anthropic critic, said the restriction was issued “reluctantly.”
The administration’s “hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release,” Sacks said.
Shutting down its top models could hurt Anthropic as it prepares for an initial public offering as soon as autumn 2026, especially if customers wander off to rival systems.
It could benefit competitors such as OpenAI, which has its own powerful cyber model and is gradually offering it to customers while talking with the Trump administration.
Friday’s moves reignited a long-standing scrap between Anthropic and the Trump administration, which remains suspicious of the startup’s liberal donor links and AI doom warnings.
Anthropic previously hired several former Biden administration officials, and the company has been fighting with the Trump administration, which has been sparring over military use of the startup’s AI tools. The Pentagon has made the unusual move of designating the company a security risk, which Anthropic is fighting in two lawsuits.
Some analysts said the White House’s dislike of Anthropic may have influenced the latest model clampdown. The senior White House official said the move was about model safety and that the Defence Department was not heavily involved.
