Chinese startup Moonshot AI is accused of stealing American technology for the Kimi K3 model.
A major new scandal is erupting in the world of artificial intelligence, directly impacting the technological rivalry between the US and China. White House AI advisor Michael Kratsios has publicly accused Chinese startup Moonshot AI of large-scale and systematic technology theft. According to him, the company secretly copied Anthropic's Fable 5 model and used it to create its flagship development — the Kimi K3 model.
Kratsios stated that Moonshot developed a specialized platform for mass distillation of models from the US. To avoid detection, the company constantly changed access routes. This is not just a single violation, but a well-organized industrial espionage operation.
Technological Distillation or Theft?
At the heart of the conflict is the distillation technique, which allows training a new AI model on the responses of a more powerful one. Open use of this technique is considered legal and widespread. However, as Kratsios emphasizes, covert distillation on an industrial scale, aimed at stealing American technology, constitutes unacceptable theft.
Anthropic confirms the allegations. According to the company's February report, specialists tracked over 3.4 million interactions with the Claude model linked to Moonshot. Traffic came from hundreds of fake accounts, and the accounts matched the public profiles of senior Moonshot employees. The attack targeted data on reasoning, coding, and image processing.
Sanctions Evasion and Hardware
Kratsios also made another serious claim: Moonshot purchased servers based on top-tier NVIDIA GB300 chips and used them in Thailand. It is worth recalling that the export of such chips to China is strictly prohibited by US regulators. Washington tightened control over AI chip exports in May, and in June, a raid was conducted at Super Micro's office in Taiwan — authorities were investigating whether chips were being smuggled into China.
Kimi K3: A New Record and Escalating Competition
Moonshot unveiled Kimi K3 on July 16. The model boasts 2.8 trillion parameters — the largest open-source AI model ever released. Demand was so high that within 48 hours, the company suspended new subscriptions. All model weights are planned to be published by July 27.
The publication date is critically important. Anthropic warns that the copied models lose their safety guardrails. Once released into the public domain, these versions could spread uncontrollably, creating serious risks.
This is not the first such incident. In January 2025, White House AI chief David Sacks accused DeepSeek of copying OpenAI models. Beijing at the time called Anthropic's findings unfounded. Now the situation is repeating with Moonshot, but this time the accusations are coming from the highest level.
My expert assessment: Kratsios has not yet announced specific sanctions, and Moonshot remains silent. However, this scandal is a clear signal that the technological war between the US and China is entering a new, more intense phase. The US still holds a massive advantage in AI investment (spending nearly 23 times higher than China), but Chinese developments are rapidly closing the gap. The next move belongs to Washington, and it could be decisive.