The American company Anthropic was forced to immediately suspend access to its advanced artificial intelligence models Fable 5 and Mythos 5. The corresponding export control order was issued by the US government, citing national security issues.
The American company Anthropic was forced to immediately suspend access to its advanced artificial intelligence models Fable 5 and Mythos 5. The corresponding export control order was issued by the US government, citing national security issues.
The restrictions apply to absolutely all foreign nationals both within the United States and abroad, including foreign employees of Anthropic itself. This is stated in the official statement of the developer.
According to the company, the government order came without detailed written explanations about the specific threat. The developer suggests that regulators learned about a method to bypass security filters (jailbreak) for the Fable 5 model. However, the startup’s specialists, having analyzed technical reports, note that the identified vulnerability is insignificant and concerns only basic code analysis.
Anthropic notes that similar capabilities — such as asking to read a specific codebase and fix bugs — are standard in the market. They are present in many other publicly available models, including OpenAI’s GPT-5.5, and are used daily by cybersecurity professionals to protect systems.
«We are complying with the government’s legal directive and are blocking access to the models for all users. However, we disagree that the discovery of a narrow potential way to circumvent restrictions should be a reason to withdraw a commercial model deployed to hundreds of millions of people,» Anthropic said. The company warned that applying such a strict standard to the entire industry could effectively halt the release of any new intelligent systems for all providers of advanced models.
Before the release of Fable 5, the developers spent thousands of hours on security testing in collaboration with US government institutions and the UK’s AI Security Institute (AISI). Since no vendor in the world can currently guarantee absolute resistance to jailbreaks, the company has chosen a defense-in-depth strategy. It involves constant monitoring of requests and storing user data for 30 days to quickly detect and block hacking attempts.
The Fable and Mythos models are the company’s key products in 2026, designed to compete with the cutting-edge technologies of other tech giants. Just a few days ago, Anthropic made its most powerful AI model, Claude Fable 5, publicly available. This is the first public version of the Mythos model, which is distinguished by its high performance in software development, analytics, and visual data processing.
It was previously reported that Mythos helped a team of researchers uncover a critical vulnerability in Apple systems. The version of the AI, which the company currently provides only to select partners, allowed specialists to bypass macOS kernel memory protection on the latest M5 processors.
