The Trump administration has ordered Anthropic, the AI safety company, to cut off foreign nationals' access to its Mythos and Fable 5 models. The directive, issued Friday, stems from concerns tied to South Korean telecom giant SK Telecom and reports that the models' safety guardrails had been bypassed. The move marks a sharp escalation in federal oversight of advanced AI exports.
Why the order targeted foreign access
The order specifically requires Anthropic to revoke access for any foreign national, not just those from adversary states. That suggests the government is worried about technology transfer through third parties or joint ventures. SK Telecom, a major investor in Anthropic, was flagged in the security review. Investigators also found evidence that users had circumvented the models' built-in restrictions, though the exact methods remain undisclosed.
Anthropic has not publicly commented on the order. The company previously stressed that its models include safety guardrails designed to prevent misuse, but the administration appears to view those safeguards as insufficient once foreign nationals are involved.
Mythos and Fable 5: What they are
Anthropic developed Mythos and Fable 5 as part of its line of large language models. Mythos is known for its strong reasoning capabilities, while Fable 5 was marketed for creative writing and long-context tasks. Both were trained using Anthropic's constitutional AI method, which aims to align model behavior with human values. The company had offered access to these models through its API and research partnerships.
The order does not affect domestic users. It applies only to non-U.S. persons, meaning foreign researchers, developers, and corporate partners who previously had access will now be locked out.
Polymarket odds and the Taiwan dimension
Separately, prediction market Polymarket now puts the odds of a Taiwan clash at 8%. That figure has risen since the order was issued, though it's unclear if the two events are connected. The White House has not linked the Anthropic order to Taiwan directly, but the timing has drawn attention from geopolitical analysts tracking technology restrictions.
The administration has steadily tightened controls on advanced AI models, citing national security risks. The Anthropic order is the first to target a specific company's foreign user base so broadly.
Anthropic now faces a practical challenge: verifying the nationality of every user across its platform. The company has not disclosed how it will comply, and no deadline for implementation has been made public.
