Overview
Three days after Anthropic launched Claude Fable 5, the most capable AI model on the market, a U.S. government order forced it offline. That withdrawal may unsettle any firm that has spent the past two years building AI tools into its work and has come to rely on them.
The problem is continuity. A team that has standardized on a single model just got a preview of what it could look like to have that model disappear for an indefinite stretch without warning. The obvious lesson is to avoid letting anything critical ride on a single tool or model.
Another wrinkle is stranger, and specific to this episode. The export-control order turned on nationality. It restricted access by foreign nationals. Anthropic, in turn, shut the model off for everyone rather than carving out a subset of users. Taking this issue to its logical extreme, global firms might have to address which lawyers and staff may use which tools, sorted by citizenship rather than role.
The directive came from the U.S. government. Reporting tied it to the Commerce Department, and to a letter from Commerce Secretary Howard Lutnick to CEO Dario Amodei. It also swept in Mythos 5, a limited-access model released the same day.
The government put Fable 5 and Mythos 5 under export control. Once something is export-controlled, providing it to a foreign national abroad is an export that needs a federal license. Less obvious is the “deemed export”: giving a foreign national access inside the United States counts as an export to their home country too. That is why nationality drove the shutdown. To keep the models available, Anthropic would have needed to somehow obtain licenses covering all affected foreign nationals accessing its new models.
The fight is unusually public. The government framed the issue as national security. According to Anthropic, officials pointed to a possible “narrow, non-universal jailbreak” that could make the model read a codebase and identify vulnerabilities by trying to fix software flaws. Anthropic says the government produced only verbal evidence, and that the same capability is “widely available from other models” including OpenAI’s GPT-5.5. It apologized to customers.
The government has its defenders. By Axios's account, Amazon warned the White House that its own researchers had jailbroken the model to reach capabilities meant to be restricted, with CEO Andy Jassy taking the concern to senior officials. The unease was bipartisan: on Face the Nation, Senator Mark Kelly, an Arizona Democrat, called the models' reach into federal and financial systems "very concerning," adding that he agreed with the administration.
At the same time, many outside experts were critical of the government action. In public posts rounded up by CyberScoop, Chris McGuire, a senior fellow at the Council on Foreign Relations, called the across-the-board restriction “highly questionable” and the deemed-export controls “just absurd.” In an open letter reported by Axios, a group of security leaders led by Alex Stamos urged the administration to lift the controls, arguing it "has taken the best models away from defenders … without any real risk to justify it."
A leading American AI lab and the U.S. government trading public shots over a flagship product is a real rift, and a strange one under an administration that has made American AI leadership a signature priority.
As of this writing, Fable 5 remains offline while Anthropic presses officials in Washington to lift the restriction. The problem it surfaced, however, may not leave with it. For a practice betting on these tools, vendor choice needs to include the continuity issue highlighted by last week’s Fable 5 drama.
The Weekly Inference is a recurring Step Into IP feature. Nothing here is legal advice.
