When a company describes its own product as “too powerful to release”, regulators pay attention. That is exactly what happened when Anthropic launched Claude Fable 5, a version of its Claude Mythos AI program. Within days, the US government ordered the company to suspend access to the model for foreign nationals, citing national security concerns. The order forced Anthropic to disable both Fable 5 and a related model, Mythos 5, for all users worldwide.
Anthropic’s Claude Fable 5 is an advanced AI system that rivals OpenAI’s ChatGPT and Google’s Gemini. The company had touted safeguards it built in to prevent misuse, but US national security authorities believed they had found a way to bypass those protections—a technique known as “jailbreaking.” Jailbreaking means getting past software restrictions designed to protect a network, allowing hackers to access sensitive information or unlock hidden features. Anthropic reviewed a demonstration of the technique and said it only identified a small number of “previously known, minor vulnerabilities” that other publicly available models could also find. The company disagreed that a narrow potential jailbreak justified recalling a model deployed to hundreds of millions of people.
“An explainer on why the US ordered Anthropic to disable its most advanced AI models and what it means for AI regulation.”
The tensions between Anthropic and the US government did not emerge overnight. Earlier in the year, Anthropic refused to allow the US military to use its AI models for domestic surveillance and fully autonomous weapons systems. In response, the government placed Anthropic on a supply chain blacklist. The latest action marks a major escalation: for years, US export controls had focused on the chips and tools that power AI, but now the government is directly restricting foreign access to the AI itself. Anthropic said it received only “verbal evidence of a potential narrow, non-universal jailbreak” and called on the government to follow principles of fair and fact-based regulation.
For UK readers, this episode has several implications. First, it highlights how quickly AI regulation is evolving and how national security concerns can overrule commercial releases. The European Union, which gained access to Mythos earlier in June after weeks of talks, said the development underlined “Europe’s need for technological sovereignty”. That means the UK and EU may accelerate their own rules on powerful AI systems. Second, the concept of jailbreaking is not new—it applies to everything from phones to software—but its application to AI models raises questions about who decides what is safe enough for public release. Finally, Anthropic’s valuation has neared $1tn (£747bn), showing that AI companies are under immense pressure to balance innovation, safety, and government relations.
Q: What is jailbreaking in the context of AI? Jailbreaking refers to bypassing the built-in safeguards of an AI model to make it do things its creators intended to block. In this case, the US government believed someone could jailbreak Fable 5 to identify software vulnerabilities—a task that, if misused, could enable cyberattacks.
Q: Why did the US order Anthropic to suspend the models? The US government issued an export control directive to stop foreign nationals from accessing Fable 5 and Mythos 5, citing national security. It did not specify exact concerns beyond a potential jailbreak method. Anthropic had no choice but to disable the models for all customers to comply.
Q: What does this mean for the future of AI development? The incident shows that governments are willing to step in decisively when they perceive a security risk, even at the cost of slowing commercial progress. It may lead to stricter global rules on who can access cutting-edge AI models and how companies must test for vulnerabilities before public release.
What happens next is uncertain. Anthropic is expected to become a public company soon, and its dispute with the US government is ongoing. The company has called for greater US oversight of AI, including a mechanism to block models with unacceptable risks—but it argues this particular order did not follow fair, evidence-based regulation. Meanwhile, the roughly 150 groups that had previewed Mythos will now get access to a version without cybersecurity or biology limitations, but only through a “trusted access program.” The broader public release of Fable 5 remains suspended indefinitely.