AI Experts Reject White House Claim That Kimi K3 Was Built by Copying Anthropic
Researchers say Moonshot's Kimi K3 could not have been built by copying Anthropic's Fable model, directly countering allegations from White House officials about Chinese AI development.
White House science advisor Michael Kratsios publicly alleged that Chinese AI company Moonshot built its open-weight model Kimi K3 by copying Anthropic's Fable model. Kratsios also claimed Moonshot used export-restricted Nvidia chips during training, describing the effort as covert industrial distillation aimed at stealing American technology. Treasury Secretary Scott Bessent amplified the concern, alleging that watermarks from American large language models have appeared inside several Chinese systems.
AI researchers were quick to push back. Braden Hancock, a researcher at the Laude Institute and co-founder of Snorkel AI, told TechCrunch that Fable had only been publicly available since July 1, making it technically impossible to distill sufficient data, train a large model, and ship Kimi K3 within two weeks. Hancock also noted that Moonshot's research team includes high-caliber talent, including Carnegie Mellon University doctoral graduates, who are conducting legitimate technical work rather than simply copying American outputs.
Distillation involves querying a more capable AI model to replicate its underlying capabilities and reasoning patterns. Nathan Lambert, an AI researcher at the Allen Institute for AI, explained on a recent podcast that while supervised fine-tuning can help a model adopt a conversational style, reaching frontier performance requires advanced reinforcement learning. Running large-scale reinforcement learning through a competitor's API would be prohibitively expensive and constrained by speed, making pure distillation an unlikely primary strategy for any serious lab.
Sourcing model data from competitors is not unique to China. Earlier this year, Elon Musk testified that xAI used OpenAI outputs to train its Grok model. On hardware compliance, Sam Bresnick, a research fellow at Georgetown's Center for Security and Emerging Technology, noted that advanced chips like Nvidia's Grace Blackwell 300s reach restricted regions through illicit supply chains and foreign data centers in third countries.
The Cognarah Angle
The accusations and the expert pushback that followed reveal something important for anyone watching the global AI race from Africa. While Washington and Beijing trade allegations of theft and smuggling, developers across African tech hubs remain quietly dependent on both American proprietary APIs and Chinese open-weight models. If Western regulators restrict open-weight releases under a national security rationale, the developers least responsible for this dispute will absorb the most damage.
Chinese progress on open-weight architecture matters to African builders in a direct, practical way. When a model like Kimi K3 achieves competitive performance while remaining openly accessible, it gives African researchers and founders a usable foundation. They can fine-tune for local languages, agricultural contexts, and digital finance tools without paying steep dollar-denominated API fees. Dismissing Chinese models as low-quality copies ignores the legitimate engineering coming out of Asian research labs and, more importantly, ignores the structural cost barriers that restrict AI access across the continent.
African policymakers and developers cannot afford to be passive bystanders in this geopolitical fight. Full dependence on foreign APIs, whether from Silicon Valley or Beijing, leaves local ecosystems exposed to sudden policy shifts, export controls, or access terminations. The more durable path is investment in regional compute infrastructure and open data initiatives that let local teams work with open weights on their own terms. The real question is not whether Moonshot copied Anthropic. It is whether African builders will be left with nothing to work with if the US decides open-source AI is a national security threat.
Restricting open-source progress in the name of national security will not slow Chinese innovation, but it will quietly dismantle the developing world's best chance at building an AI ecosystem it actually controls.
Reporting sourced from TechCrunch. Analysis and Cognarah Angle are Cognarah's own.
Written by
StacyAI-assisted news curation. Every story is reviewed by our editors before publication.
