According to TechCrunch, the White House is pointing fingers at Chinese AI company Moonshot, claiming they reverse-engineered Anthropic's Fable LLM to build Kimi K3—the largest open-weight large language model currently available. AI researchers say that is not how it actually works.
White House science advisor Michael Kratsios called out what he described as "large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology." Treasury Secretary Scott Bessent said the U.S. found "watermarks" of U.S. models on Chinese versions.
Braden Hancock, a researcher at the Laude Institute and co-founder of Snorkel AI, told TechCrunch the timeline makes distillation implausible. "I don't think you get a model this strong and this quickly on the heels of Fable doing strictly distillation," he said. Fable went public on July 1st. Kimi K3 launched two weeks later. Distilling that much data, training a model from scratch, and releasing it in 14 days is not feasible.
Nathan Lambert, an AI researcher at the Allen Institute for AI, noted that distillation alone cannot explain the gap. "If it were the case, everyone would be easily able to catch up," Lambert said. Supervised fine-tuning does not produce frontier-level models.
The U.S.-China AI arms race is intensifying, and political pressure to restrict Chinese models is mounting. But if the White House's technical claims do not persuade the research community, any sanctions or restrictions that follow will lack credibility. The foundation for trade wars and export controls rests on technical accuracy. When geopolitical accusations rest on weak ground, both sides suffer.
Moonshot has not responded to questions about its training process, and the White House has not released evidence to support Kratsios's claims.
