A powerful new AI tool has been abruptly disabled just days after its public release, because the US government fears it can be 'jailbroken' — a term that describes getting around the software's safety restrictions to exploit vulnerabilities in computer systems. The incident has put a spotlight on a growing tension between AI developers and regulators over how to assess the risks of these advanced models.
At the centre of the story is Claude Fable 5, a version of Anthropic's Claude Mythos, an AI program that rivals OpenAI's ChatGPT and Google's Gemini. Anthropic, the company behind the tool, said it self-described the model as "too powerful" ahead of its release. The company enabled pre-release access for a handful of organisations in April because the tool was so intelligent that it could be dangerous due to its ability to exploit or hack computer systems. When Fable 5 was released publicly, technology, finance and government leaders expressed concerns. Soon after, the US government ordered Anthropic to suspend access to Fable 5 and Mythos 5 for foreign nationals, citing national security concerns. Anthropic said it would "abruptly disable" the models for all customers to comply.
“Explains AI jailbreaking, why the US government ordered Anthropic to disable its models, and what it means for UK users.”
The order stems from concerns about what is known as a "jailbreak" — a method of bypassing software restrictions designed to protect a cyber network, allowing hackers to access sensitive information or unblock features. Anthropic said its understanding is that the US government believes it has become aware of a method of jailbreaking Fable 5 to identify software vulnerabilities. However, the company reviewed a demonstration of the technique and found that the vulnerabilities were "relatively simple" and could be discovered by other publicly-available models without needing a jailbreak. Anthropic also noted that the government did not provide specific details of the concern, only "verbal evidence of a potential narrow, non-universal jailbreak". The company said it disagreed that a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.
For UK readers, this matters because AI models are increasingly used by businesses and individuals across the country. If a model can be jailbroken, it could be used to hack into systems, steal data, or cause financial harm. The incident also highlights the global reach of US export controls on AI. The European Union, which gained access to Mythos earlier in June, said the development further underlined "Europe's need for technological sovereignty". The EU is assessing the situation. Moreover, the tension between AI developers and regulators is likely to shape how AI is governed in the future, potentially affecting the tools available to UK users.
Q: What is AI jailbreaking? AI jailbreaking is the process of bypassing the safety restrictions built into an AI model to make it do things it was not designed to do, such as accessing sensitive information or identifying security flaws in computer systems.
Q: Why did the US government order Anthropic to disable its AI models? The US government issued an export control directive to suspend access to Fable 5 and Mythos 5 for foreign nationals, citing national security concerns. It believed a jailbreak method existed that could allow the model to find software vulnerabilities.
Q: How does this affect users in the UK? The order forced Anthropic to disable the models for all customers globally, including UK users, to ensure compliance. The incident also raises wider questions about the security of AI tools and the potential for regulatory actions that could limit access to advanced AI.
What happens next is uncertain. Anthropic is in an ongoing separate lawsuit with the Trump administration over an order to stop government agencies using its AI tools. The company had previously called for greater US oversight of AI, but said the government action on Fable 5 did not follow principles of fair and fact-based regulation. Meanwhile, the Pentagon's chief information officer stated that some things are "simply more important than revenue cycles". Anthropic is expected to become a public company soon, with a private valuation near $1tn (£747bn). The company intends to expand access to a safer version of the model through a broader trusted access program, but for now, Fable 5 and Mythos 5 remain suspended.