U.S. Order Forces Global Shutdown of Anthropic AI Models

The Trump Commerce Department just pulled the plug on Anthropic’s two most powerful AI models for foreign users worldwide, citing national security and a risky jailbreak.

Story Snapshot

The Commerce Department ordered Anthropic to halt access to its Fable 5 and Mythos 5 models for all foreign nationals, even inside the United States, under export-control powers.^[1]^[3]
Officials pointed to a jailbreak that could bypass safeguards and expose cyber tools, but did not share detailed technical proof in writing.^[1]^[3]
Anthropic shut the models off for everyone because it says it cannot reliably verify who is a U.S. citizen, angering paying customers worldwide.^[1]^[3]
Anthropic argues the vulnerabilities are minor and common across frontier models, raising sharp questions about government overreach and selective enforcement.

Trump Team Uses Export Controls To Clamp Down On Frontier AI

Trump administration officials at the Department of Commerce used existing export-control powers to issue a formal directive against Anthropic’s newest artificial intelligence models, Fable 5 and Mythos 5.^[1]^[3] The order, sent late in the day, told the company to suspend access for foreign nationals anywhere in the world and for foreign nationals living inside the United States.^[1]^[3] That sweeping scope shows the government is treating advanced model “weights” like sensitive defense technology, not just consumer software, under the broader diffusion rule for frontier AI.^[2]^[3]

Under that diffusion framework, the Bureau of Industry and Security inside Commerce can demand licenses for export, re-export, and even some domestic transfers when it believes advanced models might flow to adversary states or help them train competing systems.^[3] The directive against Anthropic’s top-tier models was described in Associated Press and other reporting as the most significant U.S. step so far to limit access to cutting-edge AI tools.^[1]^[3] That makes this decision a test case for how far Washington will go in the name of stopping leaks to China and other hostile regimes.^[3]

Jailbreak Fears Versus Thin Public Evidence

Commerce officials grounded their order in a claimed jailbreak that could bypass safety guardrails on Fable 5 and reach deeper cybersecurity functions, including code review and vulnerability hunting.^[3] According to Anthropic’s public statement, the government said someone had demonstrated a method to get around the model’s protections and surface security flaws in software, raising fears that foreign actors could weaponize the system.^[3] Yet the letter itself, Anthropic says, did not include specific technical details or a written explanation of the threat, beyond citing national security concerns.^[1]^[3]

After the order, Anthropic conducted its own forensic review of the jailbreak demonstration that officials described. The company says the technique uncovered only a few minor, previously known software vulnerabilities and that other public models, including OpenAI’s GPT‑5.5, can find the same type of bugs without any jailbreak at all. Anthropic also points to thousands of hours of joint “red teaming” with U.S. and U.K. government experts and private groups, which it says never produced a universal jailbreak or catastrophic failure. From its point of view, the government has not shown unique, model-wide danger tied to Fable 5 or Mythos 5.

Foreign Nationals, Model Weights, And A Blanket Shutdown

The directive did not only restrict sales overseas; it also barred access by foreign nationals living in the United States, including many of Anthropic’s own employees and contractors.^[1]^[3] That “deemed export” logic comes from long‑standing rules on dual‑use technologies, where simply allowing a foreign person to use or study a system can count as an export.^[3] But here, the decision had an immediate side effect: Anthropic said it could not quickly verify which users were U.S. citizens or green‑card holders, so the company disabled Fable 5 and Mythos 5 for everyone, worldwide.^[1]^[3]

That move hit regular developers and businesses who had pre‑paid for access and who had no link to hostile governments. Social‑media creators and small firms reported lost money and stalled projects after their non‑U.S. workers were suddenly locked out. For many conservatives, this feels like a familiar pattern: broad, one‑size‑fits‑all rules that punish law‑abiding Americans and allies while bad actors overseas keep pushing ahead. The Trump administration is now being forced to walk a tightrope between real national‑security concerns and the risk of choking off innovation with red tape and sweeping nationality tests.^[3]

Anthropic Backed Strong Controls—Now Says Standard Would ‘Halt All New Models’

There is a twist that should not be lost on readers: Anthropic has been one of the loudest voices in favor of tougher export controls on advanced chips and model weights.^[2]^[4] In a detailed comment on the diffusion rule, the company wrote that “maintaining America’s compute advantage through export controls is essential for national security and economic prosperity” and urged Washington to strengthen enforcement.^[2]^[4] Anthropic effectively asked the government to treat leading AI models more like sensitive weapons technology than ordinary software, in order to keep China from catching up.^[2]^[4]

Anthropic disabled Claude Fable 5 and Mythos 5 after a national security order.
Access was cut for all users, including employees and paying customers.
The directive cited a narrow ‘jailbreak’ risk but gave no detailed explanation.
This is the first time a commercial AI product… pic.twitter.com/GZPl7pzCII

— Appify News (@appify_news) June 13, 2026

Now that its own flagship models are under the microscope, Anthropic is warning that the standard used against Fable 5 and Mythos 5 would, if applied across the industry, “halt all new model deployments.” The company stresses that it has seen no evidence of harmful real‑world misuse tied to the jailbreak and says the cited vulnerabilities are minor and common. That clash shows the deeper problem with opaque national‑security decisions: when officials will not share technical details, the public is left to choose between trusting government judgment or trusting the word of a billion‑dollar company with skin in the game. For conservatives who value both strong defense and limited government, this case will be an early benchmark for how the Trump administration handles that balance in the AI age.^[1]^[3]