The White House's distillation claim against Kimi K3 is now facing real technical pushback. If experts are right that the timeline makes distillation implausible, the sanctions threat we covered yesterday rests on a shakier foundation than the government initially suggested.
- White House science advisor Michael Kratsios claimed Moonshot built Kimi K3 by copying Anthropic’s Fable while using export-restricted chips.
- Researcher Braden Hancock said the timeline makes strict distillation implausible, since Fable has only been public since July 1 and K3 shipped roughly two weeks later.
- AI researcher Nathan Lambert said distillation has become less impactful as Chinese models approach the frontier and shift toward reinforcement learning training instead.
- Neither Moonshot nor Anthropic responded to TechCrunch’s questions about K3’s training process or Fable-specific distillation claims.
- Distillation is common industry-wide, not unique to China. Elon Musk testified earlier this year that xAI distilled OpenAI models to help build Grok.
What Happened
White House science advisor Michael Kratsios claimed Moonshot, the company behind Kimi K3, built the model by copying Anthropic’s Fable while using Nvidia chips not cleared for export to China, TechCrunch reported. Kratsios called it “large-scale, covert industrial distillation” but did not share supporting evidence, and Moonshot did not respond to questions about its training process.
Multiple AI researchers pushed back on the distillation claim specifically. Braden Hancock, a researcher at the Laude Institute and co-founder of Snorkel AI, said the timeline makes strict distillation implausible, since Fable has only been public since July 1 and K3 shipped roughly two weeks later, not enough time to distill that volume of data, train a model, and release it.
Nathan Lambert, an AI researcher at the Allen Institute for AI, said distillation has become less impactful as Chinese models approach the frontier and training shifts toward reinforcement learning. He noted that if distillation alone explained the gap, other labs would have already caught up using the same technique, and they haven’t.
Experts explained that copying a frontier model’s most advanced capabilities likely requires reinforcement learning at massive scale, tens of millions of agents grading responses, rather than simpler techniques. Running that scale of training through a frontier lab’s own API would be extremely costly, slow, and might not even produce a meaningful capability gain.
Why It Matters
This directly complicates the sanctions threat we covered yesterday. If credible technical experts believe the specific distillation claim doesn’t hold up on timeline grounds alone, the policy case for sanctions is resting on a less certain technical foundation than officials have publicly suggested.
This isn’t a case of Chinese labs being cleared entirely. Anthropic itself accused Moonshot, DeepSeek, and MiniMax of systematic distillation earlier this year, based on internal usage pattern analysis, a claim distinct from this specific Fable allegation. Distillation itself is also common practice across the industry broadly, not a uniquely Chinese behavior, Elon Musk testified earlier this year that xAI distilled OpenAI’s models while building Grok.
“I don’t think you get a model this strong and this quickly on the heels of Fable doing strictly distillation.” Braden Hancock, Researcher, Laude Institute
Bottom Line
Watch whether the White House or Treasury Department releases actual technical evidence for the Fable distillation claim, rather than restating it. Absent that, expert skepticism on timeline grounds alone is a meaningful check on how seriously to weight the sanctions threat right now.
For SaaS founders evaluating Chinese open-weight models like K3, this doesn’t resolve the underlying regulatory risk we flagged yesterday, sanctions could still happen regardless of whether this specific technical claim holds up. But it does suggest the situation is more contested and uncertain than a straightforward theft narrative implies, worth factoring into how urgently you plan around it.
