The U.S. government has ordered a sweeping ban on foreign nationals—including employees of AI research firm Anthropic—from accessing two of its most advanced AI systems, Fable 5 and Mythos 5, citing “national security concerns,” according to a company blog post issued three days after the public launch of Fable 5.
The directive, confirmed by sources familiar with the matter, marks the first time a U.S. government agency has imposed such restrictions on access to commercially developed AI models, raising questions about how closely the Biden administration is monitoring AI deployment amid rising geopolitical tensions. Anthropic, which developed the models in partnership with the Defense Advanced Research Projects Agency (DARPA), has not disclosed which government agency issued the order or the specific security risks involved.
Anthropic’s blog post, published late Thursday, stated that the ban applies to all foreign nationals—even those with work authorization—without exception. The company did not specify whether existing users of the models would be required to cease access immediately or if the restriction would be retroactive. A spokesperson for the National Security Agency (NSA), which oversees cybersecurity and AI-related threats, declined to comment on whether the agency was involved in the decision.
Industry analysts say the move reflects growing concerns within U.S. intelligence circles about the potential for adversarial actors—particularly those from China, Russia, and Iran—to exploit cutting-edge AI tools for espionage or disinformation campaigns. Last month, the White House issued a national security memo outlining stricter controls on AI research, though it did not mention specific models or companies. The new ban on Fable 5 and Mythos 5 suggests a more targeted, operational response.
Why the restriction targets foreign nationals—and not U.S. citizens—remains unclear. Some cybersecurity experts, speaking on condition of anonymity, suggest the decision may stem from concerns about data leakage or unauthorized access to training datasets, which could include classified or sensitive information. Others point to a broader trend of U.S. agencies tightening controls on AI tools used by foreign researchers, particularly in fields like generative AI and large language models.
Anthropic’s Fable 5, unveiled on May 20, was designed as a next-generation AI system capable of reasoning across multiple domains, including cybersecurity, logistics, and autonomous systems—a profile that aligns with defense applications. Mythos 5, a separate model developed under the same DARPA contract, is understood to focus on adversarial robustness, a key priority for U.S. military and intelligence agencies. The timing of the ban—just days after launch—suggests the government may have identified vulnerabilities or unintended capabilities during initial testing.
The restriction does not appear to affect other AI models developed by Anthropic, including its consumer-facing products like Claude, which remain available to users worldwide. However, the move could signal a shift in how the U.S. government regulates AI tools moving forward, particularly those with potential dual-use applications. A senior official at the Department of Defense, who requested anonymity, said the decision was made after “rigorous review” but declined to provide further details.
For now, the ban leaves open critical questions about its scope and enforcement. Unlike previous restrictions—such as those on Chinese telecom equipment or certain cryptographic tools—the order does not specify which countries or individuals are affected. Without clearer guidance, companies like Anthropic face the challenge of determining who qualifies as a “foreign national” under the directive, particularly in cases where employees hold dual citizenship or work visas.
Industry observers note that the ban could also complicate international collaboration in AI research, an area where the U.S. has historically relied on partnerships with allies like the UK, Canada, and Japan. A spokesperson for the European Commission said the bloc was “monitoring developments closely” but had no immediate comment on whether the restriction would impact EU-based researchers or companies.
The next steps for Anthropic and the U.S. government remain uncertain. The company has not indicated whether it will appeal the directive or seek exemptions for specific users. Meanwhile, the Biden administration’s AI Safety Summit, scheduled for later this month, may address broader questions about how to balance innovation with national security in the AI sector.
