OpenAI and Google Sell AI Access to Chinese Giants Through Singapore Loophole

OpenAI and Google have confirmed to Financial Times journalists that their AI services are reaching Singapore-based subsidiaries of three Chinese tech giants: Alibaba, Baidu and Tencent. All three parent companies appear on the Pentagon's so-called 1260H list, which covers companies suspected of ties to the Chinese People's Liberation Army.

A Loophole in the Rules

The case was reported by the Financial Times, which laid out the mechanism Chinese firms use to legally sidestep US export restrictions. American rules limit access to advanced AI based on geographic location rather than corporate ownership. In practice, that means a company registered in Singapore and owned by a Chinese conglomerate can freely use OpenAI's or Google's models, even though its parent company sits on the US Department of Defense's blacklist.

OpenAI told the Financial Times that it blocks direct access to its models from mainland China, but allows some firms with Chinese capital to use its services in jurisdictions where appropriate safeguards can be applied. That distinction, based on where the subsidiary is registered rather than where its capital comes from, is precisely the loophole the investigation describes.

Suspected Distillation

OpenAI had already responded to worrying signals before the investigation was published. A month earlier, the company suspended API access for users linked to Alibaba over suspected distillation, the practice of using a advanced model's outputs to train and improve competing systems. It's a practice American AI firms have been guarding against for months, viewing it as a way to sidestep their own research and development investments.

The most concrete figures in the case came from Anthropic, which told the US Congress that Alibaba had set up roughly 25,000 fake accounts used to carry out more than 28.8 million interactions with the Claude model, in violation of the service's terms. Anthropic is the only major American AI firm to have imposed an outright ban on the use of its advanced models by China-based entities, a stance that now positions the company as an industry reference point.

The Trump administration keeps saying we need to beat China in the AI race, but the problem is it hasn't done anything about export controls - Chris McGuire, Council on Foreign Relations

Sanctions Threat From Washington

The case carries extra weight in light of the latest statements from the US Treasury Department. Treasury Secretary Scott Bessent said Washington is considering sanctions against foreign AI models if they are found to be unlawfully using technology developed by American companies. Among the models named were China's Kimi K3 from Moonshot AI and DeepSeek.

If it turns out foreign models are using our major companies' technology unlawfully, we have the ability to sanction them - Scott Bessent, US Treasury Secretary

Bessent said the administration's main concern remains model distillation and unauthorized access to intellectual property. Anthropic previously secured approval to pay $1.5 billion in damages over unlawful use of copyrighted books to train its models, underscoring that IP violations in AI carry a financial dimension in Washington's eyes running into billions of dollars.

Pressure on the Commerce Department

Anthropic's hardline stance toward Chinese customers now puts the US Commerce Department under direct pressure to either close the subsidiary loophole or effectively sanction practices that OpenAI and Google have followed for months. Critics of the Trump administration, such as Chris McGuire, quoted by the Financial Times, point to a gap between the stated goal of winning the technology race against China and the actual lack of hard export-control tools that keep pace with the ways companies get around the rules.

For the AI industry, the case signals growing regulatory risk that could affect not only American model providers but also third-party firms using their APIs across Southeast Asia. If Washington decides to close the ownership loophole, providers like OpenAI and Google will have to either introduce stricter verification of corporate customers or accept the loss of revenue from a region where numerous subsidiaries of Chinese conglomerates are registered.

For Polish companies and institutions using OpenAI's or Google's models through global APIs, the case matters indirectly, as a signal that access rules for the most advanced models could tighten in the coming months, along with identity and location verification procedures for business users. Any further steps by the US Treasury and Commerce Departments could also affect the terms under which European customers use these same services.