Anthropic restricts access to advanced AI models following US government directive

Artificial intelligence company has suspended access to its latest and most advanced AI systems for foreign nationals

Artificial intelligence company has suspended access to its latest and most advanced AI systems for foreign nationals after receiving a directive from the United States government citing national security concerns.

In a statement released on Friday, the company behind the popular AI assistant Claude revealed that federal authorities had instructed it to immediately restrict access to its newly introduced AI models, known as Fable 5 and Mythos 5. According to Anthropic, the government order was delivered late Friday afternoon and required the company to prevent all non-US citizens from using the technology.

The restriction applies not only to users located outside the United States but also to foreign nationals currently residing or working within the country. This includes international employees at Anthropic who had previously been authorised to access the systems as part of their work.

Company officials said the directive was issued with little advance notice, forcing Anthropic to implement the restrictions rapidly. The firm noted that the government did not provide detailed explanations regarding the specific security risks that prompted the decision, leaving the company with only limited information about the concerns involved.

At the centre of the controversy is Mythos 5, a highly sophisticated artificial intelligence model designed with advanced cybersecurity capabilities. The system is particularly effective at identifying software vulnerabilities and security weaknesses within computer code. In some cases, the technology has reportedly detected flaws that had remained hidden for many years, enabling organisations to strengthen their digital infrastructure and improve cyber defence systems.

Government agencies and selected private-sector partners have used the technology to identify and repair security gaps before they can be exploited by malicious actors. However, experts have long debated the potential risks associated with such powerful AI tools. Critics argue that the same capabilities that make the system valuable for defence purposes could also be misused to discover and exploit vulnerabilities, potentially turning the technology into a sophisticated cyberweapon.

Fable 5, which was launched earlier this week, is built on the same underlying technology as Mythos 5. However, several of its most powerful cybersecurity and biotechnology functions have already been disabled for public users. The complete version of Mythos 5 remains available only to government agencies and a limited group of approved corporate partners.

Anthropic stated that it had reviewed reports believed to have influenced the government’s decision. According to the company’s internal assessment, the concerns appear to focus on the model’s ability to analyse computer code, identify weaknesses and suggest corrections. Company representatives argued that these capabilities are not unique and are commonly found in other advanced AI systems available today.

The firm pointed out that competing models, including those developed by major AI companies, possess similar coding and debugging functions. Anthropic therefore questioned whether restricting access to technology used by millions of people worldwide was an appropriate response.

The company also defended the safety measures built into Fable 5, stating that the system had undergone extensive testing before its release and includes safeguards designed to reduce misuse.

The latest development comes shortly after Anthropic publicly called for greater international cooperation on artificial intelligence governance. Earlier this month, the company suggested that leading AI developers should coordinate efforts to slow or temporarily pause the development of increasingly powerful AI systems if necessary.

Anthropic warned that the pace of advancement in artificial intelligence is accelerating rapidly and argued that policymakers and technology companies should retain the ability to pause development if future systems become difficult to control. The company maintains that careful oversight and global collaboration will be essential to ensuring that advanced AI technologies remain safe, secure and beneficial for society.