On 22 July 2026 Michael Kratsios, Director of the White House Office of Science and Technology Policy, stated publicly that the US government had information indicating Moonshot AI had distilled Anthropic’s Fable model in order to develop Kimi K3.
The claims made
As reported, the accusation had several parts. Kratsios alleged that Moonshot had built a sophisticated internal platform for conducting large-scale distillation against US models, and that it had rotated access methods in order to avoid detection. Separately, and on a different subject, he alleged that Moonshot had obtained GB300-equipped servers, accessing restricted Nvidia hardware by way of Thailand.
The two allegations are worth keeping apart. The chips claim, if substantiated, concerns existing export-control law with established enforcement mechanisms. The distillation claim concerns conduct for which the legal framework is considerably less settled.
The evidentiary position
Neither the White House nor Anthropic published evidence directly connecting the alleged extraction activity to Fable 5 or to Kimi K3. That gap was noted immediately and became the focus of the response from researchers, who pointed both to the absence of published evidence and to the compressed fifteen-day window between Fable 5’s return to availability and K3’s launch.
This is the central difficulty running through the whole dispute. Attribution of a model to a teacher is an unsolved technical problem. Behavioural similarity is weak evidence, because models trained on web text now routinely absorb other models’ outputs indirectly. Traffic analysis can show that extraction occurred without showing what was built from it.