U.S. Forces Anthropic to Pull Its Most Powerful AI Models Over Hacking Fears

A 5:21 p.m. letter that became a global kill switch

Anthropic received an order from the U.S. Commerce Department at 5:21 p.m. ET on Friday, June 12, instructing it to suspend all access to its two most powerful models — Claude Fable 5 and Mythos 5 — "by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees."cnbc Because the company cannot reliably sort users by nationality, the practical effect was a worldwide shutdown: within hours it pulled both models for every customer, just three days after Fable 5's public debut.lawfaremedia The directive cited "national security authorities" but offered no specifics, and Anthropic apologized to customers while calling the disruption a "misunderstanding."cnbc

A first-ever use of export-control law on AI

Commerce Secretary Howard Lutnick said he acted because officials feared the models could be diverted to military intelligence users in China, Russia or other countries of concern, according to a letter sent to CEO Dario Amodei.reuters The order threatened "prompt criminal and civil penalties" for noncompliance and marks the first time the department has used powers under the 2018 Export Control Reform Act to control an AI model.reuters The legal theory is novel and contested: export controls have traditionally not applied to software accessed remotely, and analysts note that for a model served over an API, the weights never leave Anthropic's servers.lawfaremedia One expert called the move "not well thought-out," noting it even bars Canadian and British employees at Anthropic from doing research.reuters

The jailbreak at the center of the dispute

The trigger was a reported "jailbreak." The government says a trusted partner — reportedly Amazon, a major Anthropic investor — demonstrated a way to bypass a safeguard that would otherwise stop Fable 5 from being used to find software vulnerabilities.lawfaremedia +1 Amazon CEO Andy Jassy raised the concern directly with administration officials.reuters Anthropic counters that the bypass surfaced only "minor" flaws that other public models, including OpenAI's flagship, can also find, and that it had tested Fable 5 with the government and received approval to deploy it.reuters +1

Regulation Anthropic asked for, but not like this

The clash lands at a delicate moment for the roughly $1 trillion company, which has confidentially filed for an IPO and long championed AI oversight.cnbc Days before the letter, Amodei argued frontier models should face airplane-style testing and that releases could be "blocked or reversed" if unsafe.cnbc But more than 150 executives and technical leaders signed an open letter demanding the directive be lifted, warning the rules are neither written down nor transparent.cnbc Senior Anthropic staff have met with officials almost daily; both sides say they want a fast resolution, though the company has not said when its models will return.reuters