Moonshot AI denies stealing from Anthropic as US-China AI tensions rise

China's Moonshot AI has found itself at the centre of a fresh US-China technology dispute after a Trump administration adviser accused the company of stealing intellectual property from American AI firm Anthropic. The allegation, levelled publicly this week, claims Moonshot's flagship model Kimi K3 was built by 'distilling' knowledge from Anthropic's proprietary Fable model.

But multiple AI experts have pushed back, arguing the charge oversimplifies how large language models are developed. They say Kimi K3's capabilities stem from legitimate research and openly available techniques, not theft.

The distillation accusation

The Trump adviser alleged that Moonshot AI used a process called model distillation to extract information from Anthropic's Fable without authorisation. Distillation involves using a larger, more powerful model to train a smaller one, which can then match much of the original's performance.

This technique is common in the AI industry but is ethically and legally contested when applied to proprietary models without consent. The accusation has sparked headlines in the US and UK, with the BBC and TechCrunch both reporting the claim prominently.

Moonshot AI has not yet issued a detailed response to the specific allegations. However, company representatives have previously emphasised that Kimi K3 was developed through independent research and massive computational resources.

Experts question the narrative

AI researchers who have analysed Kimi K3's architecture say the model shows signs of original design, not simple replication. Dr. Ananya Sharma, a machine learning researcher at an Indian institute, told TIVRA News that distillation leaves distinct fingerprints that are absent in Kimi K3.

"If Kimi K3 were merely a distilled version of Fable, we would expect to see specific patterns in its output distribution and internal representations. Those patterns are not there," she said. Other experts pointed out that open-weight models from Meta and others have been widely used as base architectures, which is legal under their licenses.

The controversy comes at a time when US-China tech rivalry is intensifying. Washington has been tightening controls on AI chip exports, and allegations of intellectual property theft serve to further polarise the landscape.

Indian startups urge restraint

In a separate development reported by Politico, a group of startup founders in India has appealed to the Trump administration not to restrict access to Chinese open-weight AI models. They argue that such models are vital for innovation in emerging economies.

"Cutting off Chinese open-weight models would hurt developers in India more than it hurts China," one founder told Politico, requesting anonymity. Indian startups rely heavily on affordable, accessible AI models to build applications for local markets.

The appeal highlights a growing divide between the US push for technology decoupling and the global South's need for accessible AI tools. Indian firms have been among the most active users of both Chinese and American open-weight models.

Whether the US will take concrete action against Moonshot AI remains unclear. The adviser's comments may signal a broader crackdown, or they may remain at the level of political rhetoric. For now, the AI community is watching closely, and Indian startups are hoping diplomacy prevails over confrontation.