OpenAI and Google reportedly sold AI to blacklisted Chinese firms via Singapore loophole

OpenAI and Google have reportedly supplied advanced artificial-intelligence services to Singapore-registered subsidiaries of Alibaba, Baidu and Tencent, whose parent companies sit on the Pentagon’s “1260H” list of firms tied to the Chinese military. The finding comes from a published investigation the two companies have only partly disputed.

The sales are legal, which is the point. Current US export controls restrict named entities and mainland China as a place, but not the Singapore-registered units of those same entities — a gap that lets restricted frontier models reach blacklisted groups. The models reportedly involved include OpenAI’s GPT-5.6 and Anthropic’s Mythos and Fable systems.

OpenAI said it suspended application-programming-interface access for Alibaba-linked users last month over suspected “distillation” of its model outputs, a technique for copying a model’s behavior, and reported the activity to the US government. “We don’t think nationality alone should decide access,” the company said.

Google said its AI tools remain available in markets including Singapore and Hong Kong under usage policies that forbid distillation, while acknowledging that geographic restrictions alone cannot fully stop sophisticated users from routing around controls.

Anthropic, cited for contrast but not a subject of the sales, said it bars Chinese-owned entities from its frontier models outright, and that Alibaba created roughly 25,000 fake accounts generating 28.8 million interactions with its Claude models. Alibaba has filed a court challenge to its 1260H designation.

The account rests on a single investigation, and the specific sales, model access and interaction figures have not been independently verified. Even so, the report has renewed calls in Washington for export controls on frontier AI software akin to the existing curbs on advanced chips.