The Trump administration recently softened its export restrictions on Anthropic’s advanced AI technology, restoring access to the Mythos 5 model for about 100 vetted companies and government agencies. However, Anthropic’s Fable 5 system remains under export controls, highlighting continued regulatory ambiguity surrounding frontier AI tools.

This partial easing follows a strict directive issued days earlier that barred all foreign nationals—including Anthropic employees—from accessing both Mythos 5 and Fable 5, citing national security risks related to potential misuse in cyberattacks. Commerce Secretary Howard Lutnick subsequently authorized limited usage of Mythos 5 under “appropriate safeguards” for trusted partners, signaling cautious reopening rather than full market availability.

Anthropic, which had just launched both models before the initial export ban, described the government action as a misunderstanding and pressed for broader access, especially for Fable 5. The company emphasized that Mythos 5 was designed to enhance cybersecurity, referencing Cisco’s deployment of the model to analyze over a billion lines of code rapidly and uncover more than 99 percent previously unknown vulnerabilities.

This episode spotlights the unresolved debate on whether leading AI systems should be managed as commercial software or treated as sensitive national security technology. Regulators currently lack a comprehensive framework, leaving decisions on AI export controls inconsistent and subject to abrupt shifts. This patchwork approach contrasts with the executive order issued weeks earlier by President Donald Trump, which opposed mandatory federal AI regulation and favored voluntary government reviews before broad AI deployment.

Anthropic’s CEO has advocated for AI systems to undergo rigorous technical testing and governmental auditing prior to release, akin to the safety certifications required in aviation. The ongoing regulatory uncertainty raises critical questions about the balance between innovation, security, and export control in the AI sector.