US Firms Sell AI Models To Blacklisted Chinese Companies

A recent investigation has revealed that OpenAI and Google have been providing advanced artificial intelligence (AI) services to Singapore-based subsidiaries of Chinese companies Alibaba, Baidu, and Tencent.

These companies are on the Pentagon's blacklist due to alleged ties to China's military. However, the sales are legal under current US export controls, which do not prohibit blacklisted Chinese firms from accessing American AI models through overseas subsidiaries.

Both OpenAI and Google confirmed the sales to the Singapore-based affiliates of the three companies. The parent groups of these companies appear on the Pentagon's 1260H list, which identifies entities with alleged links to the Chinese military.

The current export controls target specific entities and geographies, restricting mainland China while leaving jurisdictions like Singapore largely untouched. This means a Singapore-incorporated subsidiary of a blacklisted Chinese firm can operate under Singaporean law and enter contracts that its parent company on the mainland cannot.

OpenAI suspended API access for Alibaba-affiliated users last month due to concerns over potential distillation, a process where developers use advanced AI model outputs to improve competing systems. The company reported the activity to the US government and said it blocks direct access to its models from China but allows some Chinese-owned companies to use its services in jurisdictions where safeguards can be enforced.

Google said its AI services are available in markets including Singapore and Hong Kong, subject to usage policies. However, the company acknowledged that geographical restrictions alone are insufficient to stop sophisticated users from circumventing controls.

In contrast, Anthropic has taken a stricter approach, banning China-headquartered companies and their foreign subsidiaries from accessing its advanced models. The company has previously accused Chinese AI labs of distillation and told Congress that Alibaba used tens of thousands of fraudulent accounts to generate exchanges with its Claude models, violating its terms of service.

This report exposes a gap in Washington's efforts to slow Beijing's AI development, highlighting the need for more effective export controls. The sales of AI models to blacklisted Chinese companies have significant implications for the US-China tech rivalry and the global AI landscape.

The incident also raises concerns about the potential misuse of advanced AI models and the need for stricter regulations to prevent distillation and other forms of intellectual property theft. As the US and China continue to compete in the AI space, the issue of export controls and the sale of AI models to blacklisted companies is likely to remain a contentious issue.