Google and OpenAI Provide AI Model Access to Chinese Tech Firms via Singapore Units

A Financial Times investigation revealed that Google and OpenAI have granted access to their artificial intelligence models to Singapore-based subsidiaries of major Chinese technology companies, including Alibaba, Baidu, and Tencent. These Chinese firms are under U.S. scrutiny due to alleged ties to the Chinese military. The access was facilitated through a legal loophole, as contracts were made with subsidiaries operating in Singapore rather than directly with companies in China, making the transactions legally permissible but raising concerns in Washington.

The primary U.S. worry is that Chinese companies might leverage advanced American AI models to enhance competing AI systems, particularly through a process called "refinement," which involves using outputs from sophisticated models to train or improve other models. OpenAI confirmed it recently blocked users linked to Alibaba over fears of unauthorized use of its development interface and reported the matter to the U.S. government. OpenAI stated its models are not available within China but are accessible in regions where usage policies can be enforced and suspicious activity monitored.

Google also offers services in Singapore and Hong Kong under restrictions but acknowledges that geographic blocks are not always sufficient to prevent sophisticated misuse attempts. In contrast, the AI company Anthropic has adopted a stricter policy, banning Chinese companies and their foreign-owned affiliates from using its models altogether.