Moonshot AI Accused of Copying Anthropic's Model as US-China AI Arms Race Intensifies

Moonshot AI's recent release of Kimi K3 has sparked international concern after US officials accused the Chinese company of illegally extracting Anthropic's proprietary technology to build their own system. The allegation underscores mounting tensions in the AI race between Washington and Beijing, even as Chinese competitors gain ground with alternative strategies that US labs have largely abandoned.

The Kimi K3 Controversy#

Moonshot AI unveiled Kimi K3 this week, drawing immediate attention for its capabilities that rival leading models from OpenAI and Anthropic. However, White House director Michael Kratsios quickly challenged the company's claims, alleging that the model was created by distilling—extracting and compressing knowledge from—Anthropic's Claude 3.5. This isn't the first time Anthropic has leveled such accusations against Chinese laboratories, but the timing and the model's evident sophistication have renewed concerns about intellectual property protection in the sector.

A Strategic Divergence#

The episode highlights a fundamental divide between US and Chinese AI development philosophies. While US companies like Anthropic and OpenAI pursue proprietary, paid models, Chinese firms have increasingly embraced open-weight systems that anyone can download and modify freely. This approach mirrors Meta's earlier Llama strategy, which the company has since abandoned in favor of building its own superintelligence division—a costly pivot that has yet to yield comparable results.

Open-source advocates argue that China's transparency accelerates innovation across the entire industry; when competitors can build on each other's discoveries rather than reinventing the wheel, progress accelerates. Yet US labs remain locked in direct competition, unable to leverage their competitors' breakthroughs. Ironically, as companies burn through budgets deploying AI, many are now considering cheaper open-source alternatives—a competitive pressure Anthropic has responded to by raising prices for Claude 3.5 rather than lowering them.

Washington's Split Response#

Inside the Trump administration, officials are divided over how to respond. The Commerce Department favors export controls to restrict China's access to advanced chips, arguing this is sufficient. Others push for executive orders to prevent AI IP theft directly—though, as one analyst noted, such orders carry little force against foreign actors beyond symbolic intent. The core tension remains unresolved: whether the issue reflects a failure of current trade policy or something that policy tools cannot address.