US and Chinese AI Labs Distill from Each Other, Double Standards Controversy Emerges
Taylor Wilson
US AI companies are building commercial products on Chinese open-source models — the mirror image of what US labs accuse China of doing. The line between 'borrowing' and 'theft' is blurring, and America's rule-setting leverage is at stake.
What did Murati's new company just ship?
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released its first model Inkling this week. Its architecture is based on China's DeepSeek-V3, with post-training data generated by Moonshot AI's Kimi K2.5.
The company just closed a $2 billion seed round at a $12 billion post-money valuation. Nvidia and AMD both invested.
Yet Inkling ranks below Anthropic's, OpenAI's, and several Chinese models on the company's own benchmarks. This means → a top-tier US AI startup used a Chinese open-source model as its foundation — and still hasn't outperformed the foundation's creator.
What happened with Cursor?
Cursor, the AI coding tool valued at $29.3 billion, marketed its Composer 2 model in March as "frontier-level coding intelligence" — without disclosing the base model.
An X user identified the underlying model as Moonshot AI's Kimi K2.5 through code inspection. Cursor then acknowledged it "did start from an open-source base model," claiming roughly three-quarters of the final compute came from its own training.
In plain terms = the skeleton was a Chinese open-source model; Cursor added the muscle — but initially didn't say where the skeleton came from.
What have US labs accused China of doing?
Anthropic has accused Moonshot AI, DeepSeek, MiniMax, and Alibaba of unauthorized distillation — using a "teacher" model's outputs to train a cheaper "student" model — from its systems.
OpenAI has made similar claims against DeepSeek. Bloomberg reports that OpenAI, Anthropic, and Google have begun coordinating to stop Chinese firms from extracting results from their frontier systems.
This means → the US position is: China training on US model outputs = theft. But now US companies are building products on Chinese open-source models — the only difference is the license.
Where exactly is the double standard?
Forbes contributor Amir Husain identified the core asymmetry: China using US models is labeled "theft," while the US adopting Chinese open-source models is called "engineering practice."
This reflects a structural gap at the license level — US developers can legally use Chinese models precisely because DeepSeek and others publish open weights. ChatGPT's and Claude's top models are not open-source.
In plain terms = China left its door open; the US walked in legally. The US locked its door; China climbed through the window, and that's called stealing. The rules themselves are consistent — but using those rules to claim moral high ground is a harder sell when your own side is benefiting from the open door.
How has China responded?
At the World Artificial Intelligence Conference in Shanghai, Vice Foreign Minister Assistant Liu Bin said "certain countries are hyping the distillation issue," calling it "counterproductive."
President Xi Jinping stated that "AI development should not be a solo act by any single country."
Meanwhile, Moonshot AI's newly released Kimi K3 matches top US models on benchmarks; its market impact has been compared to the DeepSeek moment of early 2025. This means → China is pushing back on the narrative level — and narrowing the product gap at the same time.
What does this all point to?
When US developers themselves rely on Chinese open-source models to build commercial products, the line between "who is borrowing from whom" can no longer be clearly drawn.
This reflects a deeper problem: if the US wants to maintain rule-setting authority on AI, it must address an awkward fact — companies in its own camp are doing the same thing.
In plain terms = it is hard to profit from someone else's open-source work, accuse them of stealing your closed-source work, and still expect the world to see the rules as fair.
Content is for reference only, not financial advice.