AI Models – Anthropic Halts Access to Advanced Systems After US Export Order
AI Models – Anthropic has suspended access to its latest artificial intelligence models after receiving an export control directive from the United States government. The order requires the company to block access to its advanced Fable 5 and Mythos 5 systems for all foreign nationals, including employees, regardless of whether they are located inside or outside the United States.
Immediate Compliance With Government Directive
In response to the instruction, Anthropic announced that it is disabling the affected models for all users. The company said the move was necessary to comply with federal requirements, even though it does not agree with the reasoning behind the decision. The announcement was made through an official company statement published on its website.
Company Questions Basis of the Decision
According to Anthropic, authorities indicated concerns related to a potential method that could bypass certain protections built into the Fable 5 model. However, the company argued that the issue identified involved only a limited set of already known weaknesses and did not justify withdrawing a commercial system used by millions of people.
The AI firm stated that applying such a standard across the technology sector could create major obstacles for future releases. It warned that if every discovery of a narrow vulnerability resulted in model recalls, the development and deployment of next-generation AI systems could be significantly slowed across the industry.
Limited Information Provided on Security Concerns
Anthropic said government officials did not provide detailed public information regarding the national security risks behind the directive. The company explained that regulators believe they became aware of a technique capable of circumventing some of the model’s safeguards.
After reviewing demonstrations of the reported method, Anthropic concluded that it primarily exposed a small number of previously documented vulnerabilities. The company further noted that comparable publicly available AI models were also capable of identifying the same issues without requiring any special bypass mechanism.
Extensive Testing Conducted Before Launch
Before introducing the Fable family of models, Anthropic said it carried out extensive security evaluations in collaboration with government agencies, independent organizations, and internal specialists. The testing process reportedly involved thousands of hours of analysis aimed at challenging the system’s built-in protections and identifying potential weaknesses.
The company stated that these assessments were conducted alongside partners in both the United States and the United Kingdom. According to Anthropic, the results demonstrated stronger safeguards than those found in earlier generations of AI systems released to the public.
No Universal Bypass Identified
Anthropic emphasized that security researchers and testing teams have not discovered what is commonly referred to as a universal jailbreak. Such a method would allow broad circumvention of safety protections and potentially unlock a wide range of restricted capabilities.
The company maintained that despite extensive examinations by external and internal experts, no technique has been found that can consistently override the model’s protections across multiple categories of sensitive tasks. Anthropic argued that this outcome reflects the effectiveness of the security measures implemented in Fable 5.
While the company continues to comply with the export control order, it has made clear that it disagrees with the conclusion that the identified vulnerabilities warrant removing access to the model. The situation highlights the growing debate over AI security standards, government oversight, and the balance between innovation and national security concerns in the rapidly evolving artificial intelligence sector.