US Bans Foreign Nationals from Anthropic’s Fable 5 and Mythos 5 AI Models

Anthropic says a US government export-control directive has forced it to shut foreign nationals out of its Fable 5 and Mythos 5 AI models, a move that could become a blueprint for broader AI restrictions if regulators keep swinging the national security hammer.

• Directive issued: June 12
• Models affected: Fable 5 and Mythos 5
• Who’s blocked: All foreign nationals, including foreign employees at Anthropic
• Anthropic’s view: The response is too broad and based on overstated jailbreak fears

Anthropic says the US government ordered it to suspend access to Fable 5 and Mythos 5 for any foreign national, whether that person is inside or outside the United States. That includes foreign employees working at Anthropic itself. The company says it complied immediately, but it also argues the directive is a step too far and may set a nasty precedent for how advanced AI systems get regulated going forward.

For readers unfamiliar with the term, export controls are government restrictions on who can access certain technology, especially when cross-border use is involved. They’ve long been used for things like advanced chips, encryption, and sensitive industrial tech. Now AI models are getting pulled into the same bucket, which tells you plenty about how nervous governments are becoming.

Anthropic says the government cited national security authorities but did not provide detailed evidence for the move. That matters, because without specifics it’s hard to tell whether this was a measured security response or another case of officials seeing a scary headline and reaching for the regulatory flamethrower.

The company suspects the trigger may have been a reported jailbreak on Fable 5. A jailbreak is a method for bypassing a model’s built-in safety rules so it can be pushed into doing things the creator tried to prevent. Anthropic says the technique in question only exposed a small number of previously known vulnerabilities, not some magic exploit that cracked the model open like a rusty bank vault.

“The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national.”

Anthropic is pushing back hard on the idea that this justifies a sweeping restriction. The company says similar capabilities already exist in other AI systems, including OpenAI’s GPT-5.5, and argues that the government may be overreacting to a narrow issue.

“We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities.”

“…similar capabilities are already available in other AI systems, including OpenAI’s GPT-5.5.”

“We believe recalling commercial AI systems over narrow jailbreak concerns could set a dangerous precedent.”

That last point is the real friction here. If a model can be restricted because someone found a jailbreak, what happens next? Do all serious AI systems get wrapped in nationality-based access rules the moment a flaw is demonstrated? Or does the standard become, “if it can be abused, lock it down and ask questions later”? That’s the kind of policy logic that sounds decisive in a memo and looks like a bureaucratic mess in the real world.

Anthropic also says Fable 5 went through thousands of hours of testing by US government agencies, the UK AI Safety Institute, private security firms, and internal red teams. Red teaming means deliberately trying to break, trick, or misuse a system to find weaknesses before bad actors do. Anthropic says no universal jailbreak has been found and that it uses a “defense in depth” approach, meaning multiple safety layers are stacked together instead of relying on one single lock.

That’s a fair defense. It also doesn’t erase the government’s likely concern. Advanced AI systems can potentially be used for cyber abuse, malware assistance, phishing at scale, persuasion campaigns, or other nastiness that makes national security people break into a cold sweat. Governments are not inventing their worries out of thin air. The problem is that governments also love vague categories, broad powers, and rules that catch legitimate researchers in the same net as actual threats.

The restriction appears to be limited to Fable 5 and Mythos 5, not Anthropic’s other models. That narrow scope suggests this is not a blanket shutdown of the company’s entire AI stack. It also raises obvious questions: why these two models, why now, and what evidence convinced regulators to target them specifically? Anthropic says it wasn’t given enough detail to answer that publicly.

Online reaction has been split. On Reddit, some users treated the directive as a legitimate national security measure, while others saw it as a routine export-control move dressed up with extra drama. Both reactions are understandable. Frontier AI really is powerful enough to justify caution. At the same time, broad access restrictions based on vague “security” claims can become policy by blunt instrument — the government’s favorite hobby when it wants to look busy.

There’s also a larger political backdrop. Reports suggest the Trump administration may be expanding export controls on advanced AI models. If that happens, AI companies could face the same kind of geopolitical squeeze that has already reshaped semiconductor supply chains. Some version of that is probably unavoidable. The open question is whether policymakers can make those rules targeted and evidence-based, or whether they’ll just throw on more restrictions because “national security” sounds good in a press release.

That tension sits right at the heart of the debate over AI safety and AI freedom. On one side, governments worry about model misuse and foreign access to cutting-edge systems. On the other, companies and researchers worry that overbroad controls will slow innovation, block legitimate collaboration, and create a permissioned AI regime that mostly punishes the honest users while bad actors route around the tape anyway. Security theater is always easier to sell than competence.

• What did the US government order Anthropic to do?
  Suspend foreign-national access to Fable 5 and Mythos 5. Anthropic says it complied right away.
• Who is affected by the restriction?
  All foreign nationals, including foreign employees at Anthropic, whether inside or outside the US.
• Why did the government act?
  It cited national security authorities, but Anthropic says it was not given detailed evidence.
• Was Fable 5 fully compromised?
  Anthropic says no universal jailbreak was found, only a small number of previously known vulnerabilities.
• Why does Anthropic object?
  The company says the response was too broad and could set a dangerous precedent for AI regulation.
• Is this part of a bigger trend?
  Yes. It appears to fit into broader AI export-control pressure, possibly under the Trump administration.

What’s happening here is bigger than two model names and one compliance order. It’s a sign that advanced AI is being treated less like ordinary software and more like strategic infrastructure. That may be justified in some cases. It may also become an excuse for sloppy, overreaching controls that treat every foreign user like a suspected spy. Those are not the same thing, even if regulators sometimes pretend they are.

Anthropic is right to push back on lazy restrictions. Governments are right to worry about powerful AI models getting misused. The hard part — the part bureaucrats usually bungle — is drawing a line that protects real national security without kneecapping legitimate research, deployment, and competition. If they can’t manage that, we’ll get the worst of both worlds: weaker freedom, weaker innovation, and a lot more paperwork for everyone except the actual bad actors.