US companies OpenAI and Google are providing access to their advanced artificial intelligence models to Singapore-based subsidiaries of Chinese giants Alibaba, Baidu and Tencent. These Chinese corporations are officially blacklisted by the Pentagon for collaborating with the Chinese military, but loopholes in US law allow them to legally use advanced US technologies.
US companies OpenAI and Google are providing access to their advanced artificial intelligence models to Singapore-based subsidiaries of Chinese giants Alibaba, Baidu and Tencent. These Chinese corporations are officially blacklisted by the Pentagon for collaborating with the Chinese military, but loopholes in US law allow them to legally use advanced US technologies.
This was reported by the Financial Times, citing official confirmations from representatives of American AI developers and their own sources in the industry.
According to the investigation, such transactions remain legal due to the specifics of US export controls. Washington’s current restrictions focus on the supply of physical chips for AI training and regulating access to some of the most powerful models, such as Anthropic’s Mythos and Fable or OpenAI’s GPT-5.6. At the same time, broader access to cloud software for companies headquartered in the PRC, even those included in the 1260H defense list, is still not prohibited at the legislative level.
Following a formal request from journalists, OpenAI said it had suspended access to its API for users affiliated with Alibaba last month over concerns about «distillation,» a process in which third-party developers use responses from U.S. systems to train their own competing AI models.
«We would prefer that most of the world use AI shaped by democratic values, not AI controlled by autocratic governments,» OpenAI commented on the situation, noting that access should not be determined solely by nationality, but the company only works in countries where it is able to control preventive measures.
Google also confirmed that its AI services are available in Hong Kong and Singapore, subject to a usage policy that prohibits distillation. However, the company acknowledges that geographical restrictions on sales are ineffective against sophisticated attackers who can easily circumvent such barriers.
Instead, Anthropic takes a much tougher stance, completely blocking access for companies with Chinese capital anywhere in the world. The other day, Claude developers eliminated technical loopholes that allowed Chinese firms to bypass restrictions. Anthropic has previously accused Chinese AI labs DeepSeek, Moonshot, and MiniMax of stealing technology through distillation. In a letter to the US Congress, the company reported that Alibaba used 25,000 fake accounts to generate more than 28.8 million messages with the Claude model to train its own systems.
