White House blocks Anthropic Mythos exports over suspected China access
Export controls on Anthropic's Mythos model were triggered by intelligence suggesting a China-linked group may have obtained the weights, Semafor reports.
The White House imposed export restrictions on Anthropic's Mythos model after intelligence indicated a group with ties to China may have accessed the system, according to a Semafor report published this week. The move marks one of the first times the U.S. government has acted on suspected foreign access to a frontier AI model before a formal breach confirmation.
Mythos 5 and its sibling model Fable 5 represent Anthropic's most capable unreleased systems. Neither has been made publicly available, and Anthropic has not disclosed technical specifications or training details. The company declined to comment on the Semafor report when reached by The Verge.
Export controls and scope
The White House decision reportedly hinged on classified assessments that a China-linked entity had obtained some level of access to Mythos, though the exact nature of that access—whether full weight downloads, API endpoints, or internal documentation—remains unclear. Export controls typically apply to model weights above certain capability thresholds, but the administration has discretion to act on national security grounds even for unreleased systems.
The restrictions prevent Anthropic from sharing Mythos weights or providing access to foreign nationals from designated countries without a license. The policy does not affect Claude, Anthropic's publicly available model family, which remains accessible globally under existing terms of service.
