Moonshot could face US sanctions: Govt takes note of Chinese AI firm's distillation attacks against Anthropic to improve its Kimi model

US artificial intelligence firm Anthropic has been accusing Moonshot AI of engaging in distillation activities. Now, the US government is threatening sanctions against the Chinese AI company, in yet another indication of the two nations' race for tech dominance. Michael Kratsios, Director of the White House Office of Science and Technology Policy, accused Moonshot of undermining American research. The irony is that Moonshot AI was widely praised for drastically reducing the cost of AI compute through open-source technologies. Its release of the massive 2.8-trillion-parameter Kimi K3 model is raising a few questions: did it reach that milestone by exploiting American AI research through model distillation? Here is what you should know.

What did the White House officials say about Moonshot AI?

In a post on X, Michael Kratsios said, "We have information that Moonshot AI distilled Anthropic's Fable for the development of its K3 model. To do this, they developed a sophisticated internal platform to conduct large-scale distillation against US models, allowing them to quickly switch between multiple methods of access to avoid detection."

Kratsios alleged that Moonshot AI had acquired Nvidia GB300-equipped servers, including access to GB300 systems in Thailand, likely to train its AI models. "Large-scale, covert industrial distillation aimed at stealing proprietary US technology and undermining American research is unacceptable," he added.

US Treasury Secretary Scott Bessent said he did not support the “theft of US IP” while hinting at possible sanctions and Entity List designations against companies found to be engaging in distillation. "We support open-source AI and the innovation it unlocks. But open source is not open season on American IP. When PRC [People's Republic of China] firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations are on the table," Bessent posted on X.

What is a distillation attack?

Distillation is the process of training a smaller or less capable AI model using the outputs generated by a more advanced model. While model distillation is a legitimate machine learning technique when conducted with permission or on proprietary models, it becomes controversial when companies use fraudulent accounts, proxy services or other methods to repeatedly query proprietary AI systems without authorisation to replicate their capabilities.

According to Anthropic, such industrial-scale distillation enables competitors to acquire advanced AI capabilities much faster and at significantly lower cost, effectively bypassing years of research and development. The company argues that this amounts to intellectual property theft when carried out without permission.

Anthropic's allegations against Moonshot AI

In a February blog post, Anthropic claimed that it had "identified industrial-scale campaigns" by Chinese AI firms Moonshot AI, DeepSeek and MiniMax to illicitly extract the capabilities of its Claude models to improve their own AI systems.

The company alleged that the Chinese AI labs generated more than 16 million exchanges with Claude through approximately 24,000 fraudulent accounts, violating Anthropic's terms of service and regional access restrictions.

Specifically regarding Moonshot AI, Anthropic claimed that around 3.4 million exchanges targeted agentic reasoning and tool use, coding and data analysis, computer-use agent development, and computer vision capabilities.

Kimi K3: Moonshot AI's most advanced model

Kimi K3, the 2.8-trillion-parameter open-weight model from Moonshot AI, was released around 16 July. It has narrowed the gap with leading US AI models and outperformed some rivals on coding benchmarks. It was found to be weaker than Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 models on several advanced evaluations.

Anthropic had alleged that Kimi relied on hundreds of fraudulent accounts spanning multiple access pathways to conduct distillation attacks. According to Anthropic, the use of varied account types made the campaign harder to detect, while metadata associated with the activity matched the public profiles of senior Moonshot AI staff.

The company further claimed that Moonshot AI adopted a more targeted approach, attempting to extract and reconstruct Claude's internal reasoning traces rather than relying solely on output imitation.

Moonshot AI is growing strongly financially

Despite the controversy surrounding its alleged use of stronger AI models, Moonshot AI has continued to grow rapidly and is reportedly considering a Hong Kong initial public offering by the end of the year.

The company reportedly reached around $300 million in annual recurring revenue in June, up from $200 million in April 2026 and $100 million in March, driven by paid subscriptions and application programming interface or API usage for its Kimi models.

Moonshot AI has raised up to $4 billion in total funding across multiple investment rounds. It is currently valued at around $31.5 billion and is reportedly aiming for a final pre-IPO funding round at a $50 billion valuation in August.