Enterprise

Anthropic's Claude Fable 5 returns after six-day government shutdown — and enterprise AI just learned a new risk

A Commerce Department export-control letter forced Anthropic to globally disable Fable 5 and Mythos 5 on June 12. Access is back, but the precedent isn't going anywhere.

By Asher Glick · Enterprise software desk · 2026-06-19 · 3 min read

Photo: Unsplash / Tim Mossholder — The U.S. Capitol building in Washington, D.C.

At 5:21 p.m. ET on June 12, Anthropic received a letter from Commerce Secretary Howard Lutnick instructing it to disable Claude Fable 5 and Claude Mythos 5. The order was narrowly framed as an export control covering foreign access, but because Anthropic can’t segregate API traffic by nationality, the practical effect was a global shutdown that lasted roughly six days.

Fable 5 had launched three days earlier as the general-availability sibling to Mythos 5, which Anthropic was already running through Project Glasswing, a controlled-access program limited to about 50 vetted organizations. The trigger, per Wall Street Journal reporting, was Amazon CEO Andy Jassy, who told Treasury Secretary Scott Bessent and other administration officials that Amazon researchers had used a prompt chain to coax Fable 5 into producing information useful for cyberattacks. Anthropic says its own review of the demonstration surfaced only a small number of previously known, minor vulnerabilities that publicly available models can find without any bypass.

The irony is structural. Anthropic’s brand is safety, and CEO Dario Amodei recently published an essay advocating for “serious and binding regulation of AI,” including the ability to block models deemed unsafe. The Commerce Department then deployed something close to that mechanism against Anthropic’s own flagship release, on a record Anthropic itself disputes.

The Pentagon had already laid the groundwork in early March, designating Anthropic a “supply chain risk” and barring military and defense contractors from using its models on government contracts. Anthropic is challenging that designation in federal court. Now Commerce has shown it can reach the commercial product too, despite a deployment Anthropic argues touches hundreds of millions of people.

For enterprise buyers, the lesson is concrete. Model availability used to be a vendor-uptime question; CrowdStrike, Microsoft, Google, Apple, and Amazon all manage that. It’s now also a question of which way an export-control letter goes on a given Friday afternoon. A safety pitch built around the right to pull unsafe models works until the regulator agrees, and pulls yours.

Sources