According to a Financial Times investigation published on Thursday, OpenAI and Google have acknowledged that they are providing cutting-edge AI models to Singapore-based subsidiaries of Alibaba, Baidu, and Tencent, even though the parent companies are listed on the Pentagon’s list of Chinese military companies.
The disclosures highlight a weakness in Washington’s attempts to impede Beijing’s advancement in artificial intelligence. The selling of cutting-edge AI processors to Chinese-owned companies overseas is restricted by U.S. export regulations, but there are currently no comparable limitations on access to AI models, enabling blacklisted companies to employ American AI services through foreign subsidiaries.
How the Gap Operates
OpenAI told the Financial Times that while it prohibits direct access to its models from mainland China, it allows some Chinese-owned businesses to use its tools in countries such as Singapore, where the service is permitted. Google confirmed a similar setup. According to U.S. regulations, which concentrate export restrictions on hardware rather than AI model access, the transactions are now lawful.
The American AI models are used by the Chinese companies to benchmark their own systems and, in certain situations, to incorporate capabilities developed in the United States into their products. Despite Washington’s recent strong efforts to plug loopholes in its broader technology export restriction regime, the arrangement still stands.
Increasing Export Control Pressure
The revelation coincides with growing criticism of loopholes in American AI regulations. Due to their ties to Chinese state businesses, Alibaba, BYD, and Baidu were added to the Pentagon’s list of “Chinese military companies” in June, increasing it from 134 in 2025 to 188. A different loophole that had permitted Chinese businesses overseas to buy cutting-edge Nvidia and AMD chips without permits was closed by guidelines released by the Commerce Department in late May.
Legislators from both the Republican and Democratic parties have submitted legislation on Capitol Hill with the goal of strengthening export regulations for AI. Bipartisan dissatisfaction with what one politician referred to as “a massive screw-up” in the current structure was detailed in a Politico piece this week.
What’s Next
There is currently a direct political challenge to the legal divide between AI model access and AI hardware. Model-access limitations are still mostly unregulated, despite the fact that chip constraints have been gradually strengthened. This loophole, according to critics, undermines the goal of the larger containment effort. An already complicated geopolitical confrontation now has a legal component because to Alibaba’s separate lawsuit against the US government to be taken off the Pentagon’s blacklist.