OpenAI and Google have both confirmed they supplied advanced AI services to Singapore-based units of three major Chinese tech companies. 

Those companies are Alibaba, Baidu, and Tencent, and all three sit on a Pentagon watch list. The list flags firms the U.S. government believes have ties to China’s military.

The story broke on Friday, July 10, thanks to reporting from the Financial Times. It has already incited debate about whether America’s AI rules have any real teeth.

Legal Deals 

Here’s the twist; none of this breaks the law. 

Current U.S. rules don’t stop Chinese-headquartered companies from using American AI tools, as long as the access happens outside mainland China. 

Singapore counts as outside China and so does Hong Kong. That means a company can sit on a Pentagon blacklist and still sign up for premium AI access, so long as it does so through the right subsidiary in the right country. 

It’s a loophole built right into the system.

OpenAI and Google 

Both companies came clean when the Financial Times asked. OpenAI told reporters it blocks direct access to its models from mainland China. 

But it does allow certain Chinese-owned businesses to use its tools in places where it believes it can keep an eye on things.

That watchful eye paid off last month. OpenAI suspended API access for users linked to Alibaba after spotting signs of something called distillation. 

Distillation happens when a developer uses one AI model’s answers to train and improve a rival model. Think of it like copying someone’s homework, then claiming it as your own original work.

OpenAI reported the suspicious activity to the U.S. government right away.

Google gave a similar explanation. Its AI products stay available in markets like Singapore and Hong Kong, and those markets follow Google’s usage rules, which forbid distillation. 

Still, Google admitted that geography alone can’t stop a determined user. A skilled operator can find ways around location-based walls.

Anthropic

Anthropic, the maker of the Claude AI models, has taken a much stricter method by banning Chinese firms entirely from using its most advanced models. 

That ban also covers any overseas companies that Chinese firms own. Anthropic has gone further to push Washington to expand export controls on AI software. 

It wants restrictions that look more like the ones already placed on advanced computer chips.

Anthropic has also pointed fingers at specific Chinese AI labs. The company has accused DeepSeek, Moonshot, and MiniMax of using distillation to build competing systems. 

In a letter sent to Congress last month, Anthropic claimed Alibaba created roughly 25,000 fake accounts. 

Those accounts, Anthropic says, generated more than 28.8 million interactions with its Claude model, breaking the company’s terms of service.

Alibaba has not responded to those specific claims. But the company hasn’t stayed quiet either. Alibaba recently asked a U.S. federal court to remove it from the Pentagon’s blacklist altogether.

The company called its inclusion on the list “arbitrary and capricious.”