Artificial intelligence company Anthropic announced on Friday that it would be forced to abruptly disable access to its most advanced AI models, Fable 5 and Mythos 5, after receiving a directive from the US government ordering the suspension of access for all foreign nationals. According to the company, the order was issued on national security grounds, although officials reportedly did not provide detailed information explaining the specific threat that prompted the action.
Anthropic stated that the directive was issued under export control regulations and required the company to prevent foreign nationals from accessing the two recently launched models. The company said its understanding is that government officials believe a method exists to bypass, or "jailbreak," certain safeguards built into Fable 5. These safeguards were designed to prevent the model from being used for sensitive cybersecurity activities, including identifying software vulnerabilities that could potentially be exploited by malicious actors.
The move comes at a sensitive time for Anthropic, which had recently shown signs of repairing its strained relationship with sections of the US government. Earlier tensions emerged after the company reportedly refused to allow its AI technology to be used for domestic surveillance activities and fully autonomous weapons systems by the US military. Following that dispute, Anthropic was placed on a government supply-chain blacklist, a measure scheduled to take effect later this year.
The government's latest action represents a significant escalation in efforts to limit access to advanced artificial intelligence technologies. Traditionally, US export controls have focused primarily on restricting access to advanced semiconductor chips, hardware, and supporting technologies that enable AI development. However, the latest directive shifts attention directly toward restricting access to the AI models themselves, marking a potentially important change in policy.
Anthropic criticised the decision, arguing that the government had presented only verbal evidence regarding what it described as a narrow and non-universal jailbreak vulnerability. The company maintained that the alleged issue did not justify disabling a commercial product currently used by hundreds of millions of people. According to Anthropic, isolated vulnerabilities should not automatically result in the withdrawal of an entire AI system from public access.
The dispute highlights a growing divide between AI developers and regulators over how risks associated with advanced AI systems should be evaluated and managed. Ironically, only days before receiving the directive, Anthropic had publicly advocated for stronger government oversight of artificial intelligence and had supported the idea of restricting models that present unacceptable risks. Nevertheless, the company argued that Friday's order failed to follow principles of transparent, evidence-based, and fair regulation.
The Pentagon's Chief Information Officer, Kirsten Davies, defended the government's position in a post on social media, stating that national security considerations must take precedence over commercial interests. She argued that protecting American interests was more important than concerns about revenue, publicity, or company valuations.
The development comes as Anthropic moves closer to becoming a publicly traded company. The firm confidentially filed for an initial public offering last month, placing it ahead of rival OpenAI in the race to enter public markets. The timing of the government order has therefore drawn considerable attention within both the technology and investment communities.
Earlier this week, Anthropic introduced Claude Fable 5, a new model representing what the company describes as a "Mythos-class" level of capability. The system was launched with extensive safety measures designed to prevent misuse in high-risk fields such as cybersecurity. Anthropic acknowledged that some users had criticised these restrictions as excessively broad, but maintained that they were necessary to ensure responsible deployment.
Experts have warned that AI systems with Mythos-class capabilities could significantly enhance the effectiveness of sophisticated cyberattacks if they fall into the wrong hands. Industries such as banking, which rely on large networks of interconnected and often decades-old technology systems, could be particularly vulnerable to misuse of such advanced AI tools. These concerns have contributed to growing regulatory scrutiny surrounding frontier AI models.
Anthropic stated that it had worked closely with US government agencies and other stakeholders before releasing Fable 5 and argued that competing AI systems from other providers demonstrated similar abilities to identify minor software flaws. The company, therefore, expressed concern that applying this standard universally could effectively halt the deployment of future advanced AI models across the entire industry.
As a result of the directive, Anthropic said it has no choice but to disable Fable 5 and Mythos 5 to remain compliant with government requirements. The company clarified that access to its other AI models would remain unaffected. It also stressed that it believes a misunderstanding has occurred and that it is actively working with authorities to restore access to the models as quickly as possible.
Further complications emerged when Amazon's cloud computing division, AWS, disclosed that Anthropic had requested the removal of access to the affected models for all users in every region. A US government official later confirmed that the Commerce Department had indeed issued an export control directive requiring the suspension of access to Fable 5 and Mythos 5 for foreign nationals.
Former White House official Dean Ball, who helped develop the administration's AI Action Plan in 2025, suggested that the implications of the directive could extend far beyond users located outside the United States. According to his interpretation, the order may effectively restrict all non-American citizens from accessing Anthropic's newest AI models, even if they are physically located within the United States. He added that users may eventually be required to verify their citizenship before being granted access.
The policy could also create unusual challenges for the company itself. Several prominent figures associated with Anthropic, including co-founder Chris Olah, AI researcher Andrej Karpathy, and philosopher Amanda Askell, were born outside the United States. While their current citizenship status has not been publicly confirmed, questions have emerged regarding whether similar restrictions could affect foreign-born personnel involved in developing the company's technology. Anthropic declined to comment on that possibility, leaving uncertainty about the broader consequences of the government's unprecedented order.
