The decision by the US government to cut off foreign access to Anthropic’s most capable AI models was set in motion by Amazon Chief Executive Officer Andy Jassy, according to a report by the Wall Street Journal.
Jassy had raised concerns directly with senior US officials including Treasury Secretary Scott Bessent, and this triggered the block, according to people familiar with the matter cited by the Wall Street Journal.
After assessing the concerns raised, US officials concluded that restricting access to the AI models for foreign governments, businesses, and individuals was the most direct course of action.
According to the WSJ report, Amazon CEO, Jassy told US officials that Amazon researchers had used a series of prompts to get Anthropic’s Fable 5 model to provide information that could aid cyberattacks, information the model was supposed to withhold.
White House officials subsequently held a meeting to determine how to respond to this revelation, and security researchers began independently testing Amazon CEO’s claims.
Officials then asked Anthropic to either fix the vulnerabilities or take the model offline, according to administration officials cited by the Journal.
The officials eventually concluded that the most direct way to manage the risk was to block foreign governments, companies, and individuals from accessing the model entirely.
President Trump signed off on the action despite reservations from within the White House about its potential to hinder AI innovation, a senior official said.
The decision was reinforced by a call on Friday between administration officials and Anthropic CEO Dario Amodei.
Some officials said the company indicated it was unwilling to work with government security experts to address the issue. This response, according to people familiar with the matter, deepened existing doubts within the administration about whether Anthropic could be trusted to manage the security risks posed by its newest models.
The administration had reportedly held this view of Anthropic for some time, viewing the company as one of the leaders in the US AI race, but one whose approach to security oversight had not satisfied US officials even before this incident.
In response to the directive, Anthropic said it was shutting off access to both Mythos and Fable for all users globally to ensure compliance, a move that could hamper efforts by companies worldwide using the tools to identify software vulnerabilities in their own systems.
The company noted that many of its researchers are foreign-born, meaning the government’s directive effectively prevents those researchers from working on Anthropic’s latest models altogether.
Anthropic had earlier disputed the technical basis for the US government’s directive. The company argued that the vulnerability identified was a narrow, non-universal jailbreak capable of surfacing only minor, previously known issues, comparable to capabilities already available in other publicly accessible models, including OpenAI’s GPT-5.5.
Earlier, Nairametrics reported that Anthropic has urged developers of advanced AI systems to create a coordinated and verifiable framework for slowing or temporarily halting AI development if future models begin improving themselves faster than society can safely manage the associated risks.
