New Advances in AI Models Bring Fresh Accusations of Chinese Theft

OpenAI and Anthropic saying they’re getting ripped off. It’s not a new accusation, or unique to this sector.

One of China’s preeminent artificial intelligence (AI) companies has trained its latest chatbot model on advanced U.S. semiconductor chips – “Blackwells” made by Nvidia – despite export controls barring their use in that country. That’s according to a Trump administration official who reiterated to Reuters that U.S. policy is “we’re not shipping Blackwells to China.”  

That news came hours after Anthropic, a San Francisco-based AI company, accused three Chinese companies – including DeepSeek, the firm accused of utilizing the Blackwell chips – of unfairly harvesting data from its own AI tech to train up their own systems.

And it comes about two weeks after OpenAI, the builder of ChatGPT, told the U.S. House of Representative’s Select Committee on China that DeepSeek is using OpenAI’s models to train up its own. “DeepSeek’s next model (whatever its form) should be understood in the context of its ongoing efforts to free-ride on the capabilities developed by OpenAI and other US frontier labs,” reads the company’s memo to the committee.

The rise of AI is a complex issue that most Americans meet with a mixture of ambivalence and unease. We’re really all over the place on the topic. We’re rightly worried about the profound effect it could have on the conditions of work as it’s applied across the economy (our own polling shows concern about what AI will mean for manufacturing jobs). We prioritize AI safety and data security over its development but we also say it’s important the U.S. be on AI’s cutting edge. Companies, meanwhile, are making major investments to do just that: Facebook’s parent company is spending $135 billion. Google’s is spending $185 billion. Amazon is spending $200 billion.

They’re racing against innovative, well-capitalized Chinese rivals, who in some AI categories – like the large language models that Anthropic and OpenAI produce – aren’t that far behind!

These accusations aren’t new. Some researchers argue that things like data harvesting and “distillation” are exactly what AI’s evolution has been built on. Maybe so. But that doesn’t account for broken Blackwell export controls. And it must be noted that Chinese state-organized intellectual property theft has been well documented across various industries for 25 years.

It’s intrinsic to Chinese industrial policy, and now it appears to be happening in an industry where the outcomes could greatly affect economic growth and millions of working lives.