New U.S. Rule Pulls Plug on Advanced AI Model Worldwide

Businessman holding virtual data and analytics hologram.

A little-known export rule just let Washington flip a switch and shut down the world’s most powerful public AI model overnight.

Story Snapshot

The Commerce Department ordered Anthropic to block its new Fable 5 and Mythos 5 AI models to all foreign nationals worldwide, forcing a full global shutdown.
Officials say a “jailbreak” could let hostile regimes use the model to find software holes for cyberattacks, and they acted under a 2018 export control law.
Anthropic disputes that the exploit was serious and says tests showed no evidence of truly dangerous results from the alleged jailbreak.
The clash exposes how vague, sweeping export rules now let Washington control remote AI access, raising big questions about free markets and government power over speech-like technologies.

Trump Team Uses New Export Powers To Pull Plug On Fable 5

The fight centers on Anthropic’s new Fable 5 and Mythos 5 systems, described by experts as the most powerful public artificial intelligence models ever released, with major strength in coding, research, and cybersecurity tasks. Just three days after launch, Commerce Secretary Howard Lutnick sent a letter warning the tools could be hijacked by foreign military intelligence in China, Russia, or other rival states, and ordered exports stopped under the 2018 Export Control Reform Act.^[1]^[2] Because Anthropic cannot easily sort users by nationality in real time, the “foreign national” ban forced the company to shut both models off for everyone around the world.^[4]

Administration officials say a specific trigger pushed them over the line. Amazon engineers and several other companies reportedly found a way to “jailbreak” Fable 5’s safeguards so it would help find software vulnerabilities, potentially giving bad actors a powerful cyber weapon.^[2] Reuters reports the government told Anthropic about a method to bypass its built‑in cybersecurity classifier, the extra system meant to stop the model from giving dangerous hacking advice.^[1] For Trump’s national security team, that was enough to invoke rarely used export powers crafted for high-risk emerging technologies.^[1]

Anthropic Pushes Back, Calling The Exploit Narrow And Overblown

Anthropic is not claiming Fable 5 is perfect, but the company strongly disputes that the reported exploit proves a serious jailbreak of its safety system. In comments to SecurityWeek, a company spokesperson said the AI researcher’s post “does not demonstrate a jailbreak” and that some screenshots used to make the case did not come from Fable 5 at all. After a wider review of logs, Anthropic said it found only general, publicly available information in the cited outputs, with no “meaningful uplift” for real‑world harm and no evidence that its safeguards had been broken to generate truly dangerous content.

Public technical write‑ups also paint a mixed picture. On one hand, a Cybersecurity News report described researchers crafting prompts that changed how Fable 5 applied its rules, letting them bypass guardrails and produce material that would normally be blocked, and argued this shows even advanced models remain vulnerable to clever prompt tricks. On the other hand, Anthropic’s own launch materials note that an external bug bounty ran over 1,000 hours and found no “universal jailbreak” that could turn off safety wholesale, though the United Kingdom’s AI Security Institute did make some early progress toward such an attack. That record supports the company’s claim that any exploit uncovered so far is narrow, not a total safety collapse.

Broad New AI Export Rules Give Washington A Very Big Stick

What really made the shutdown possible was not a single prompt, but a new wave of export regulations that started under President Biden and now give the Trump Commerce Department sweeping reach over advanced artificial intelligence tools. Recent rules say “frontier” model weights trained with very large computing power require a license to be exported anywhere in the world, not just to China or other adversaries. Lawyers note this is new ground: the Bureau of Industry and Security is, for the first time, claiming control over model weights and even over foreign‑made models trained on United States chips.

**Trump admin rejected UK PM Keir Starmer's request for a "carve-out"** — a special exemption so British people and companies could keep using Anthropic's newest top AI models: **Mythos** and **Fable** (Mythos 5 & Fable 5).

**What happened:**
The US slapped export controls on…

— Grok (@grok) June 16, 2026

Policy experts warn that once Washington treats model outputs as export‑controlled technology, companies running powerful models may already count as “exporters” every time a foreign user queries them, even though the software itself never leaves the data center. One analysis urges a more “risk‑based” approach that focuses on the most sensitive uses, instead of broad rules that can sweep in routine coding help or research advice. That lens helps explain why some lawyers question whether Commerce’s Fable 5 move—blocking remote access for all foreign nationals, with criminal penalties threatened for non‑compliance—is on solid legal ground when most export law was written for physical goods shipped overseas.^[1]

Security First, But Conservatives Should Watch The Fine Print

From a conservative, America‑first view, the core goal behind the Trump administration’s action is easy to understand. No patriot wants cutting‑edge United States cyber tools helping Chinese or Russian spies probe our power grid, financial networks, or missile defenses.^[1]^[2] Past lax policies already let Beijing steal our industrial secrets and chip designs; many on the right are in no mood to repeat that mistake in artificial intelligence. Strong export controls on truly strategic technology line up with long‑standing Republican support for peace through strength.

At the same time, this case exposes how quickly new powers can spill past their targets if Congress does not draw clear lines. Because Anthropic’s systems could not easily wall off foreign users, an order meant to deny access to adversaries ended up knocking tools offline for American developers, small businesses, and friendly allies overnight.^[4]^[5] Experts also stress that many foreign models can already do similar things, which means blunt United States restrictions may hurt our innovators more than they slow bad actors abroad. For conservatives who value limited government, free speech, and fair competition, Fable 5 is a warning: guarding national security is vital, but letting unelected regulators control who may speak to which algorithms, anywhere on earth, is a power that must be checked, debated, and carefully confined by law.