The government provided no specific technical explanation for the restriction.
In a moment that reveals the deepening friction between sovereign authority and the ambitions of artificial intelligence, Anthropic was compelled by a US government directive to suspend two of its most capable newly released models — Fable 5 and Mythos 5 — on national security grounds, with no technical explanation offered. The order arrived days after the models' public launch, severing access not only for customers worldwide but for Anthropic's own foreign national employees. It is a reminder that the most consequential boundaries in the age of AI are not always drawn by engineers, but by governments — and that the rules governing this frontier remain contested, opaque, and still being written.
- A Friday government directive gave Anthropic no room to negotiate: comply with export controls or face legal consequences, forcing the company to pull Fable 5 and Mythos 5 offline within hours of receiving the order.
- The suspension landed with particular sting because the models had only just launched — days of public availability erased, along with the engineering investment and momentum behind what Anthropic had called state-of-the-art achievements.
- The roots of the concern trace back to Claude Mythos Preview, an earlier system whose advanced cybersecurity capabilities had already alarmed regulators and Wall Street alike, making these successor models a natural target for scrutiny.
- The restriction swept broadly — cutting off customers in Europe, researchers in Canada, startups in Japan — suggesting the government's anxiety extends well beyond military applications, even as no specific threat was articulated.
- Anthropic pushed back publicly, calling the process opaque and unfair, and demanding that any security-based AI restrictions be grounded in statute, technical evidence, and transparent procedure — standards it says this directive failed to meet.
- The episode deepens an already litigious relationship: Anthropic is actively suing the Trump administration over a separate DOD supply chain risk designation, and this latest directive adds new pressure to an unresolved and escalating standoff.
On a Friday, Anthropic received a government directive with the force of export control law behind it: disable Fable 5 and Mythos 5, or face consequences. The company complied within hours. Every user lost access — customers, partners, and even Anthropic's own foreign national employees. The government cited national security, but offered no specific technical explanation for why these two models, above all others, posed a threat.
The timing sharpened the blow. The models had launched just days earlier to considerable fanfare. Fable 5 marked the first time Anthropic had released such a capable system to a general audience, made possible by new safeguards against high-risk outputs. Mythos 5 built on similar foundations. Both were gone almost as soon as they had arrived.
The concern had earlier roots. Claude Mythos Preview, the system underlying both models, had drawn government and investor attention when it appeared in April, distinguished by sophisticated cybersecurity capabilities that regulators viewed with wariness. Anthropic had been cautious with it from the start, limiting access to a select group through a program called Project Glasswing rather than releasing it broadly.
In its public response, Anthropic expressed frustration not just with the outcome but with the process. The company argued that security-based restrictions on AI should be transparent, technically grounded, and rooted in clear statutory authority — and that this directive met none of those standards. It also apologized to its customers for the disruption.
The suspension arrives amid an already fractured relationship with the US government. Earlier this year, the Department of Defense designated Anthropic as a supply chain risk — a label historically applied to foreign adversaries — after negotiations between the two sides broke down. Anthropic responded with a lawsuit against the Trump administration that remains unresolved. The breadth of the current restriction, which affects commercial users globally rather than just defense-adjacent ones, suggests the government's concerns run deeper than any single use case. Anthropic's other Claude models remain available, pointing to a narrowly targeted action — but one whose rationale, for now, remains unexplained.
On Friday, Anthropic received a directive from the US government that would force the company to make an immediate choice: disable two of its newest artificial intelligence models or face consequences for violating export control law. The company chose to comply. Within hours, Fable 5 and Mythos 5 were taken offline for every user—customers, partners, and even Anthropic's own foreign national employees. No exceptions. The order cited national security authorities, though the government provided no specific technical explanation for why these particular models posed a threat.
The timing was striking. Fable 5 and Mythos 5 had only just been released to the public days earlier. Anthropic had announced them as significant achievements, models that performed at state-of-the-art levels across multiple industry benchmarks. Fable 5 was notable as the first time the company had released such an advanced model to a general audience, made possible by new safeguards designed to prevent the system from generating responses in high-risk domains. The suspension meant that investment, engineering effort, and the public rollout were all halted almost as soon as they had begun.
The two models were built on the foundation of Claude Mythos Preview, a system that had already drawn scrutiny from both Wall Street and government officials when it appeared in April. That earlier model had demonstrated sophisticated cybersecurity capabilities—the kind of technical prowess that can cut both ways in the eyes of regulators. Anthropic had been cautious about it from the start, declining to release it broadly and instead offering access only to a select group of companies participating in what the company called Project Glasswing, a cybersecurity-focused program.
In a statement released Friday, Anthropic expressed frustration with the process itself. The company said the government had not explained the specific national security concerns that prompted the directive. More broadly, Anthropic argued that if the government wanted to restrict AI deployments on security grounds, it should do so through a process that was transparent, fair, technically grounded, and rooted in statutory authority. This action, the company said, did not meet those standards. Anthropic also apologized to its customers for the disruption.
The suspension is the latest chapter in an escalating conflict between Anthropic and US government agencies. Earlier this year, the Department of Defense had designated Anthropic as a supply chain risk following failed negotiations between the two sides. That designation—historically reserved for foreign adversaries—requires defense contractors to certify that they will not use Anthropic's Claude models in military applications. The company responded by filing a lawsuit against the Trump administration, seeking to overturn the blacklisting. That case remains in litigation.
What makes the current suspension notable is that it affects all users, not just government or defense-related ones. A customer in Europe, a researcher in Canada, a startup in Japan—all lost access to models they may have been building products around. The breadth of the restriction suggests the government's concern extends beyond military or sensitive government use cases. Yet without specifics, it is difficult to assess whether the caution is warranted or whether it represents an overcorrection. Anthropic's other Claude models remain available, which suggests the government's concern is narrowly focused on these two particular systems. The company has not indicated whether it plans to challenge the directive or seek to modify it through further negotiation.
Citas Notables
The government should have the ability to block unsafe deployments through a transparent, fair, clear, and technically grounded statutory process— Anthropic statement
La Conversación del Hearth Otra perspectiva de la historia
Why would the government move so quickly to disable models that had just been released? What changed between launch and Friday?
The government didn't say. That's part of what frustrated Anthropic. The directive cited national security but offered no technical specifics. It's possible the models were flagged during internal review, or perhaps the Mythos Preview's cybersecurity capabilities raised alarms that carried over to these new versions.
But Anthropic says it built safeguards into Fable 5 specifically to prevent high-risk outputs. Doesn't that suggest the company was already thinking about these concerns?
It does. Which is why Anthropic's complaint about the process rings true—they were trying to be responsible, and the government still shut them down without explaining why the safeguards weren't sufficient. That's the transparency gap.
The order affects foreign nationals, even Anthropic's own employees. That's a pretty blunt instrument.
It is. It means a French engineer working at Anthropic's US office can't access the models. A researcher collaborating from Canada is locked out. It's not a targeted restriction on exports; it's a blanket prohibition. That suggests the government sees the risk as fundamental to the models themselves, not just to where they might end up.
How does this fit into the larger DOD conflict?
It's part of a pattern. The DOD already designated Anthropic as a supply chain risk. Now the government is using export controls to restrict specific products. Anthropic is fighting the DOD designation in court, but this new directive is a separate legal authority. It's like the government is approaching the problem from multiple angles simultaneously.
What happens next for Anthropic?
That's unclear. The company could challenge the directive, seek clarification, or try to negotiate. But they're already in litigation with the administration over the DOD designation. Adding another legal fight over export controls could be costly and time-consuming. Meanwhile, customers who were counting on these models have to find alternatives.