Anthropic suspends Claude Fable 5 and Mythos 5 by order of the US government over national security concerns

On June 13, 2026, the BBC published a high-impact report: Anthropic, the company behind the AI assistant Claude, has been forced to suspend two of its most advanced models —Claude Fable 5 and Mythos 5— after receiving an order from US national security authorities. The measure came just days after the public launch of Fable 5, at a time when the company was already embroiled in a legal battle with the Trump administration.

The immediate trigger for the suspension was a government order requiring Anthropic to prevent foreign nationals from accessing Claude Fable 5. However, to ensure compliance with that order, the company decided to go further and completely disable both models for all of its customers, both domestic and international. In its public statement, Anthropic wrote: «The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all of our customers to ensure compliance». It is a radical measure that underscores the severity of the regulatory pressure the company faces.

The technical reason cited by US authorities revolves around a «jailbreaking» technique, that is, a method to circumvent the safety restrictions built into the model. Curiously, Anthropic itself played down the magnitude of the threat: after reviewing a demonstration of the technique in question, the company stated that the method made it possible to identify «a small number of minor and previously known vulnerabilities», and added that those same vulnerabilities can be discovered by other publicly available models without the need for any bypass. This raises an uncomfortable question: is the suspension a proportionate response to a real technical risk, or are there other political motivations at play?

The political context is inseparable from the technical issue. The relationship between Anthropic and the Trump administration had been openly contentious during the months leading up to this episode. Donald Trump publicly criticized the company, and then-Defense Secretary Pete Hegseth went so far as to label Anthropic a «supply chain risk», a designation historically reserved for companies based in countries considered adversaries of the United States. For an American company to receive this label for the first time in history is an unprecedented and extraordinarily significant development in terms of industrial and security policy.

The «supply chain risk» designation has immediate practical consequences: it implies that the tool or service is not considered secure enough for government use. Anthropic responded by suing the Pentagon. And the courts sided with it, at least provisionally: a US judge ruled that the Pentagon's directive could not be enforced, meaning that government agencies and organizations working with the US military can continue to use Anthropic's tools while the litigation continues. However, the suspension of Fable 5 and Mythos 5 represents a new front in that confrontation.

Claude Fable 5's trajectory up to its suspension is itself revealing of the tensions surrounding the development of frontier AI models. In April 2026, Anthropic enabled private pre-launch access for a handful of selected organizations, on the grounds that the model was so powerful it could be dangerous, given its capabilities to exploit or hack computer systems. The company itself self-described it as «too powerful to release» before going ahead and launching it publicly on June 9. This apparent contradiction was criticized by some industry observers, who interpreted the statement as marketing hyperbole rather than a sincere technical assessment.

What does seem beyond doubt is the model's offensive capability in the field of cybersecurity. Professor Gina Neff, Chair of Responsible AI at Queen Mary University of London, cited data from the UK government's AI Security Institute: in its tests, the model was able to exploit defenses and systems 73% of the time. «It is a qualitative leap in cybersecurity capabilities», Neff said. This figure, if confirmed, effectively represents a danger threshold very different from that of previous generations of models, and would justify —at least partially— the authorities' concern.

Nonetheless, Professor Neff also warned about the negative side effects of restricting access. According to her, limiting who can use and test these models hampers the safe development and evaluation of the AI systems themselves, as well as restricting collaboration with governments around the world in the field of security. «We are in uncharted territory at this point», she declared. «People within the AI industry have warned us that these tools are improving very quickly and that we need to be able to develop the capabilities to keep our companies safe from cyberattacks». The paradox is evident: suspending access to powerful tools may provide short-term protection, but it may also slow collective preparedness against threats that are evolving just the same.

From Europe, the reaction was swift and laden with geopolitical symbolism. The European Union had gained access to Mythos in early June 2026, after weeks of negotiations. The spokesperson for the European Commission, Thomas Regnier, noted that the new episode underscores «Europe's need for technological sovereignty», and confirmed that the EC was assessing the situation. That same month, the European Commission had presented concrete measures to reduce the 27 member countries' dependence on US and Asian technology, including artificial intelligence. The suspension of Fable 5 and Mythos 5 thus comes at a time when Europe is actively seeking its own alternatives and reinforcing its narrative of strategic autonomy in AI.

In broader terms, this episode illustrates several structural tensions that define the current moment in the development of advanced AI. First, the tension between transparency and security: Anthropic chose to publicly announce that its model was «too powerful», which generated both legitimate anticipation and skepticism. Second, the tension between innovation and regulation: the offensive cybersecurity capability of models like Fable 5 raises genuine questions about whether current regulatory frameworks —national and international— are adequate to manage this new generation of tools. Third, the geopolitical tension: the fact that a private American company receives the same designation historically applied to companies from adversary countries is a precedent that could have lasting implications for the technology industry.

For the agentic AI ecosystem, this case is especially relevant. Frontier models with advanced autonomous capabilities —able to exploit systems, navigate complex environments and act on behalf of users with minimal human oversight— represent precisely the kind of tool that AI agents require. If the regulatory response to these advances is suspension and access restriction, the implications for the development and deployment of agentic AI are significant: more friction, more legal uncertainty, and potentially a fragmentation of the global market between jurisdictions with different permissiveness thresholds.

The article also mentions that the BBC contacted the US Department of Commerce for comment, with no response on record at the time of publication. The absence of a detailed public explanation from the US government about the «specific vulnerabilities» that prompted the order —Anthropic noted that authorities did not identify concrete concerns— adds opacity to a process that is already complex and politically charged.

In short, the suspension of Claude Fable 5 and Mythos 5 is not just a computer security incident: it is an episode that condenses into a single event the major fractures of the present moment in AI —between companies and governments, between the United States and Europe, between openness and control, and between the pace of technological innovation and the capacity of institutions to manage it—. Closely following how the litigation between Anthropic and the Pentagon is resolved, and what decisions the European Commission makes on technological sovereignty, will be fundamental to understanding the framework in which agentic AI will operate in the coming years.