Tech industry voices pressed U.S. government officials this week over potential security vulnerabilities in Anthropic’s most advanced artificial intelligence models, a person familiar with the discussions told Reuters. Among those who raised concerns was Amazon CEO Andy Jassy, the source said.
The disclosure came as the U.S. administration on Friday issued an order directing Anthropic to block use of its latest models, Fable 5 and Mythos 5, by any foreign nationals regardless of whether they are inside or outside the United States. Anthropic responded by announcing it would disable access to those models worldwide.
In a company blog post, Anthropic said the U.S. government believes there is a technique to bypass - or "jailbreak" - a safeguard intended to stop Fable 5 from being used to identify software vulnerabilities. Anthropic described the government action as an export control in that post.
The Commerce Department’s Bureau of Industry and Security, the agency responsible for export controls, did not immediately reply to a request for comment, the company said.
Observers who generally support export limits on advanced AI were nonetheless puzzled by the administration’s directive because it extends restrictions to allied countries as well as adversaries. "This was not well thought-out," said Jimmy Goodrich, a senior fellow at the University of California’s Institute for Global Conflict and Cooperation. "It even bans Canadians and Brits employed at Anthropic from doing research and development."
The order arrived amid signs that an earlier dispute between administration officials and Anthropic - which has confidentially filed for a U.S. initial public offering - had been easing across parts of the federal government. The new directive represents a sharp intervention by U.S. authorities into who may access the company’s cutting-edge models.
Amazon did not immediately respond to requests for comment about the company’s engagement with government officials or the administration action.
The developments highlight tensions at the intersection of national security and commercial AI development: government officials citing potential misuse and a private company moving to restrict access worldwide in response to regulatory pressure. The situation underscores uncertainty for AI developers, international researchers, and cloud and software security sectors as policymakers and companies navigate how to control distribution of powerful models.