US Orders Anthropic to Disable Claude Fable 5 and Mythos 5 Over National Security Concerns

Anthropic has abruptly disabled access to its most advanced AI models, Claude Fable 5 and Claude Mythos 5, after the US government issued a national security directive blocking all foreign nationals from using them.

The order marks one of the most dramatic government interventions yet in the rollout of frontier AI models, and it may signal a new phase in how powerful artificial intelligence systems are regulated, exported and controlled.

According to Anthropic, the US government directed the company to suspend access to Fable 5 and Mythos 5 for any foreign national, whether they are inside or outside the United States. The restriction also applies to foreign national employees working at Anthropic itself. Because the company could not immediately separate access by nationality in a safe and compliant way, it disabled both models for all users.

That is what makes this moment so significant.

This is not a routine product pause, a server outage or a normal safety review. It is a government-ordered shutdown of frontier AI access based on national security concerns, and it affects models that had only just begun reaching users and selected partners.

Anthropic says the directive arrived suddenly and did not include detailed evidence explaining the specific national security concern. The company believes the government’s concern may relate to a method of “jailbreaking” Fable 5, allowing the model to identify a small number of known and relatively minor software vulnerabilities.

Anthropic disputes the scale of the threat. The company says the vulnerabilities involved were simple, previously known and discoverable by other publicly available AI models. It also says there is no evidence of a universal jailbreak that broadly bypasses Fable 5’s safeguards or unlocks a wide range of dangerous cyber capabilities.

That disagreement is now at the centre of a much bigger debate.

Fable 5 was designed as Anthropic’s most powerful generally available model, bringing what the company has described as Mythos-class intelligence to ordinary users with heavy safeguards. Mythos 5, by contrast, was the more restricted version intended for trusted partners, security teams and controlled use cases. Together, they represented Anthropic’s attempt to commercialize extremely capable AI while managing the risk of misuse.

Now, the US government appears to have decided that the risk is still too high at least for foreign access.

The move immediately raises questions about whether advanced AI models are beginning to be treated more like strategic national assets than ordinary software products. For years, Washington’s AI controls focused heavily on chips, especially restricting China and other countries from accessing the most advanced semiconductors needed to train and run frontier models. This order suggests the regulatory focus may be expanding from hardware to the models themselves.

That shift could have enormous consequences.

If governments begin restricting access to frontier models based on nationality, AI companies may be forced to redesign how they deploy products globally. They may need citizenship checks, residency controls, export licences, customer verification systems and internal access restrictions for staff. That would turn AI deployment into something closer to controlled defense technology than consumer software.

It also creates a major problem for global companies.

AI labs are international businesses with international employees, customers, researchers and enterprise partners. A rule that blocks foreign nationals from using a frontier model does not just affect users in rival countries. It can also affect engineers, researchers, customers and partners inside the United States who are not US citizens.

That is one reason Anthropic pushed back so strongly.

The company says it supports government authority to block genuinely unsafe AI deployments, but argues that such decisions should follow a transparent, technically grounded and fair process. In this case, Anthropic says the government’s action was based on limited evidence and risks setting a standard that could freeze frontier model launches across the industry.

That argument will resonate across Silicon Valley.

Every major AI company knows that its models can potentially be misused. OpenAI, Google, Meta, xAI and Anthropic all face the same tension: the more capable models become, the more useful they are for legitimate work and the more dangerous they may become in the wrong hands.

Cybersecurity is the clearest example. A powerful model can help defenders find and patch vulnerabilities before attackers exploit them. But the same capability can also help attackers discover flaws faster. That dual-use nature is exactly why Mythos-class models have attracted intense attention from governments, banks, cloud providers and security agencies.

Anthropic’s own position is that Fable 5’s safeguards are among the strongest ever deployed, even to the point that some users complained they were too broad. The company says it worked with the US government, the UK AI Safety Institute, private third parties and internal teams to red-team the model for thousands of hours before release.

Read more: Anthropic’s Mythos AI Heads to UK Banks as Cybersecurity Fears Escalate

But the government clearly remained unconvinced.

The result is a model recall unlike anything the AI industry has seen at this scale.

For users and developers, the immediate impact is simple: Fable 5 and Mythos 5 are offline until Anthropic and the US government resolve the dispute. Other Anthropic models remain available, meaning Claude users are not completely cut off. But the company’s most advanced systems are now effectively locked behind a national security fight.

For the AI industry, the impact is much larger.

This could become the moment when frontier AI entered the export-control era.

Until now, the public conversation around AI safety has focused heavily on hallucinations, bias, copyright, job displacement and misinformation. But this incident pushes another issue to the front: whether the most advanced models are powerful enough to be treated as controlled strategic technology.

That is a very different kind of debate.

If AI models can accelerate cyber operations, biological research, weapons development or intelligence analysis, governments will not treat them like normal SaaS tools. They will treat them as capabilities with national security implications. And once that happens, the open global rollout model that helped AI products scale so quickly may begin to fracture.

There is also a competitive risk.

If US companies face strict controls while foreign or open-source models remain available elsewhere, American AI labs could find themselves constrained in ways that rivals are not. Anthropic has already warned that applying this standard across the industry could halt new frontier deployments. That may sound dramatic, but it reflects a real concern: safety regulation that is too reactive or inconsistent could weaken the companies it is meant to govern.

At the same time, the government’s concern cannot be dismissed entirely.

AI models are becoming more capable every month. The gap between a helpful coding assistant and a system that can meaningfully assist cyber operations is narrowing. If officials believe a model can be jailbroken to expose sensitive capabilities, they may feel pressure to act before the risk spreads.

That is the uncomfortable balance now facing the industry.

Move too slowly, and dangerous capabilities may spread. Move too aggressively, and innovation could be chilled by emergency interventions based on incomplete evidence.

Anthropic’s shutdown of Fable 5 and Mythos 5 shows that this balance is no longer theoretical. It is already shaping product access, company operations and global AI competition.

And it may mark the beginning of a new reality: the most powerful AI models will not simply be launched.

They will be cleared, controlled and contested.