Government Kills Anthropic’s Top AI

The federal government just forced one of America’s top AI companies to shut down its most powerful model worldwide in the name of “national security,” and even U.S. customers are feeling the squeeze.

Story Snapshot

The U.S. Commerce Department ordered Anthropic to block foreign use of its new Fable 5 and Mythos 5 AI models, citing security risks.^[1]^[3]
To obey the order, Anthropic disabled the models for everyone, including American users and even its own foreign-national staff.^[1]^[2]
Officials say a jailbreak could expose serious cybersecurity tools inside Mythos 5, while Anthropic calls the risk narrow and the order a “misunderstanding.”^[1]^[3]
The move shows how fast-growing export controls on advanced tech can spill over into everyday users and raise concerns about government overreach.^[1]^[3]

What Exactly Did the Government Do to Anthropic’s New AI?

The United States Commerce Department sent Anthropic a late-day directive telling the company it could not let any foreign national use its newest models, Fable 5 and Mythos 5, anywhere in the world.^[1]^[2]^[3] The order covered people overseas and foreign nationals living inside the United States, even those working at Anthropic itself.^[1]^[2] Because the company said it could not cleanly separate users by citizenship, it chose to shut both models off for everyone to avoid breaking the rule.^[1]

Anthropic had just launched Fable 5 publicly, describing it as a powerful assistant built on top of an even more advanced core model called Mythos 5.^[3] Fable 5 could run software tasks, help with cybersecurity work, and act more autonomously than earlier Claude models.^[3] After the directive arrived, Anthropic pulled Fable 5 and Mythos 5 offline while leaving its older Claude models, such as Opus 4.8, available to users.^[1] That meant regular American customers suddenly lost access to the company’s most capable tools overnight.

Why Did Officials Say These Models Are a National Security Risk?

According to reports, the Commerce Department acted after learning that another company had found a way to “jailbreak” Mythos 5.^[1]^[3] That jailbreak allegedly let users bypass some of Fable 5’s guardrails and tap into advanced cybersecurity functions inside Mythos that Anthropic normally kept locked down.^[1] Officials worried this kind of access could reveal software vulnerabilities or give dangerous actors powerful hacking help, and they used export control powers that normally target dual-use technologies to stop foreign access quickly.^[1]^[3]

Export controls are usually meant to keep sensitive tools, like advanced chips or cutting-edge encryption, out of the hands of hostile regimes and foreign adversaries. In this case, the government applied that same logic to a general-purpose AI system before any public proof of harm had surfaced.^[3] Supporters of the move argue that with models this strong, waiting for an actual breach would be reckless, especially when the tools might be used to attack American networks, power grids, or military systems from abroad.^[3]

Anthropic Pushes Back and Calls the Shutdown a “Misunderstanding”

Anthropic did not hide its frustration. In its official statement, the company said it is complying with a “legal directive” but argued the action is not “transparent, fair, clear, and grounded in technical facts.”^[3] Leaders at the company say the jailbreak described by the government was limited to a narrow scenario, not a wide-open back door that destroyed all safety protections in Fable 5.^[1] They insist they already shipped these models with strong safeguards and are now working to restore access as soon as they can.^[3]

The US government just banned the world's most powerful AI model.

Not China. Not Russia. Britain. Canada. Australia. Everyone.

Yesterday at 5:21pm the Commerce Department sent Anthropic one letter.

By midnight Fable 5 and Mythos 5 were offline globally.

No warning. No… pic.twitter.com/zwvMrO05Ip

— Kuroyami (@kuroyamiai_) June 13, 2026

The shutdown also landed in the middle of a broader legal fight between Anthropic and the Trump administration over how far federal agencies can reach into private deployment decisions for artificial intelligence.^[2] The company says it accepts that the government must be able to pause unsafe deployments, but it wants a clear process written in law, with public standards and technical evidence, rather than sudden letters that pull products with little explanation.^[3] That tension puts conservative concerns about due process and limited government squarely in the middle of this AI debate.

What This Means for Conservatives, Innovation, and Government Power

For many on the right, this episode cuts both ways. On one hand, national security is not a game, and no serious person wants hostile regimes using cutting-edge American AI to probe our defenses. On the other hand, when a Washington agency can flip a switch and shut down a major U.S. product worldwide in hours, it raises real fears about creeping bureaucratic power. Today the target is cybersecurity features; tomorrow it could be speech, politics, or gun-related tools.

President Donald Trump recently signed an executive order that lets the federal government review the highest-risk AI systems for up to a month before public release.^[3] Supporters say this keeps America ahead of enemies like China and Iran while still letting innovators move fast. Critics warn that vague “risk” language and secret evidence can become a blank check for unelected officials. For conservatives who care about free enterprise and constitutional limits, the Anthropic case is an early test of how that balance will work in practice.

How This Fits a Bigger Pattern in High-Tech Controls

This fight over Fable 5 and Mythos 5 mirrors earlier battles over advanced microchips, encryption tools, and even biological research. In each case, the government used export rules to block access based on what a technology could do, not just what it had already done.^[3] The same playbook is now being used for artificial intelligence, where a single model can write code, scan systems, and help users hunt for weaknesses in ways that could be used for good or for harm.

For everyday Americans, the risk is that broad national security claims become the default reason to shape which tools they can use, even inside the United States. For innovators, sudden export actions can chill investment and push cutting-edge work overseas. The Trump administration now faces a hard task: protect the country from real threats linked to frontier AI, while also proving to voters that these powers will not turn into yet another excuse for permanent, unelected control over American technology and speech.