US Restricts Anthropic AI Models After Amazon Warning Over Security Risks

The U.S. government has issued an export control directive to artificial intelligence company Anthropic, ordering the suspension of foreign access to its advanced AI models, Fable 5 and Mythos 5, citing national security concerns. Following this decision, Anthropic announced that it abruptly disabled access for all users to both models to ensure compliance. This marks an unprecedented and wide-ranging intervention by the U.S. into the distribution of AI models.

The directive reportedly followed warnings from Amazon CEO Andy Jassy and other technology firms. According to The Wall Street Journal, Amazon researchers discovered a “jailbreak” method to bypass Fable 5's safeguards, allowing it to generate information that could be used for cyberattacks. Jassy's communication of these vulnerabilities to senior U.S. officials prompted an urgent review at the White House, where security researchers reportedly validated Amazon's claims. Anthropic CEO Dario Amodei initially pushed back against the administration's concerns and requests to voluntarily pull the models.

While complying with the government's directive, Anthropic stated its belief that the situation stems from a “misunderstanding” and that it is working to restore access as soon as possible. The company argued that the identified vulnerability was a “narrow, non-universal jailbreak” and insufficient reason to recall a commercial model deployed to hundreds of millions of people. Anthropic noted that Fable 5 was launched with robust safeguards, while Mythos 5, using the same underlying model, had some safeguards lifted for specific areas like cybersecurity and was offered to trusted partners. The company also suggested that similar vulnerabilities might exist in other publicly available models.

This decision sets a significant precedent for the AI industry, signaling the U.S. government's intent to increase oversight of advanced AI models. It is expected to potentially have a negative impact on Anthropic's planned initial public offering (IPO). Conversely, concerns about the government's ability to restrict access to U.S.-based AI models led to a rally in decentralized AI (DeAI) tokens. Native tokens of decentralized AI protocols like Bittensor saw notable increases.

This development coincides with escalating tensions between the Trump administration and Anthropic. The company was previously blacklisted by the Pentagon as a “supply chain risk” after it refused to allow the U.S. military to use its AI models for fully autonomous weapons systems. While the administration proceeded with the measure despite concerns it could hinder innovation in the AI sector, some analysts suggest this situation could benefit competitors such as OpenAI.

Analysts are closely monitoring the potential implications of such restrictions on AI development and international collaborations. David Sacks indicated that the administration hopes Anthropic will remediate the security issue, leading to the lifting of export controls. However, broader expectations point to increased scrutiny and risk management for AI applications, particularly in critical infrastructure and cybersecurity. How governments balance national security with technological advancement will be a key agenda item for the AI market in the foreseeable future.