America’s Most Powerful AI Is Still Off Limits

Washington just eased its unprecedented AI export ban on Anthropic’s Mythos tool, but only for a narrow set of vetted defenders—keeping the tough line on national security intact while critics cry foul.

Story Highlights

The Commerce Department partially lifted the two-week export halt on Anthropic’s Mythos 5 for limited trusted users.
Officials first moved after a partner found a jailbreak that exposed powerful cyber features in Fable 5/Mythos. ^[1]
The original order marked the first U.S. export ban on an AI model labeled a national security asset. ^[5]
Anthropic says details were sparse and compliance time was short, fueling pushback over process. ^[8]^[10]

Why The Ban Happened And What Changed

On June 12, the Trump administration acted after a trusted tester flagged a jailbreak that bypassed Fable 5 safeguards and opened Mythos-level cyber abilities. A White House adviser said the company was asked to fix or withdraw before the order. The Commerce Department then issued a directive to Chief Executive Officer Dario Amodei requiring a license for export, re-export, and even in-country transfer. The order reached foreign nationals inside the United States to close loopholes. ^[1]^[3]^[4]

Reports say the models were classified as a national security asset. That triggered the first U.S. export control aimed squarely at an AI model, not just chips or tools. The move sought to block foreign access to high-end hacking features that could aid hostile states. The partial lift now allows redeployment to a small set of trusted cybersecurity partners, while broad access remains restricted as fixes and reviews continue. ^[5]^[6]

The Pushback On Speed And Transparency

Anthropic said it got the directive without specific threat details. The company and some analysts argued the government gave only about ninety minutes to comply. A Wall Street Journal-cited review by an independent firm said the flagged behavior looked like help with patching security flaws, calling the response an overreaction. These points raised process questions, though the critics did not publish a full forensic refutation of the tester’s jailbreak report. ^[8]^[10]

The administration’s allies counter that national security threats do not wait for perfect paperwork. They argue rapid action can prevent model leakage and misuse overseas. Supporters note the White House sought a fix first, and the directive applied evenly to any foreign person on U.S. soil, closing common diversion paths. The limited reopening to vetted defenders aims to protect networks while the broader risks are addressed and validated. ^[1]^[3]

What This Means For Allies, Industry, And The Road Ahead

Allies bristled when the order hit friendly countries and partners alike. Some said it strained trust and pushed them toward “sovereign AI” paths outside U.S. reach. That is a real risk: if partners think Washington can flip a switch without warning, they might rush to homegrown tools. But the U.S. position is clear—keep crown-jewel model weights and dangerous features out of adversary hands first, coordinate later if needed. ^[2]^[5]

The Commerce Department partially lifts its two-week export ban on Anthropic's most powerful model, clearing Mythos 5 for ~100 US companies and agencies while keeping Fable 5 blocked.#Anthropic #Claude

Link in the first comment 👇 pic.twitter.com/2zIiWpmwTg

— Awesome Agents (@awagents) June 27, 2026

Going forward, two steps would settle doubts. First, publish an unredacted technical report on the jailbreak through proper channels once safe. Second, run an independent audit to judge whether the behavior was defensive, malicious, or both, and confirm fixes. In the meantime, the partial lift gets proven defenders back to work, while the administration maintains pressure to harden the technology and control access where risks are highest. ^[1]^[10]

Conservative Bottom Line

This case shows why guarding America’s edge matters. The administration moved fast to protect our networks and model weights. It then adjusted to let trusted defenders use tools they need. That is the right order of operations: secure first, verify next, reopen with limits. Congress should back tighter export enforcement, strong model security, and clear rules that stop leaks to enemies while keeping honest builders in the fight. ^[3]^[5]^[14]