OpenAI, Google provided AI to Singapore units of Chinese firms

OpenAI and Google confirmed they provided advanced artificial intelligence services to Singapore‑based subsidiaries of Alibaba, Baidu and Tencent, companies the US government has linked to China’s military. The disclosures have renewed debate in Washington over whether export controls on advanced AI models should be tightened.

OpenAI said it suspended API access for Alibaba‑affiliated users last month after detecting suspected “distillation,” a process where outputs from one model are used to train or improve another. The company reported the activity to US authorities and said it does not permit access to its models from within mainland China. OpenAI noted it allows some firms with Chinese ownership to use its tools in countries where it can enforce safeguards and monitor for misuse. “We would rather see more of the world using AI shaped by democratic values than AI controlled by autocratic governments,” the company added.

Google confirmed its AI services remain available in Hong Kong and Singapore under the company’s usage policies, which include restrictions against distillation. The company warned that geographic blocks alone cannot fully eliminate risk because sophisticated users may try to circumvent location‑based controls.

Anthropic has taken a stricter approach, barring Chinese companies and foreign entities owned by them from accessing its most advanced models. The company acknowledged enforcement has been challenging, said it closed loopholes that previously allowed some Chinese‑linked entities to bypass safeguards, and has accused several Chinese AI developers of engaging in distillation. Anthropic also told lawmakers that Alibaba used thousands of fraudulent accounts to make millions of exchanges with its Claude model, an activity the company says violated its terms of service.

US export rules currently restrict high‑end semiconductors and impose controls on specific frontier AI models, including certain systems from Anthropic and an identified version of OpenAI’s GPT. Those rules do not broadly bar Chinese‑headquartered firms, including some on the Pentagon’s 1260H list of entities alleged to have ties to the People’s Liberation Army, from accessing advanced AI software through overseas affiliates.

Companies supplying the services say the transactions complied with law and that contracts and monitoring measures are in place to detect and block suspected misuse. Some lawmakers and national security experts have called for broader export limits on advanced AI models similar to existing chip controls. Debate in Washington is focused on whether contractual and technical safeguards can prevent techniques such as distillation and whether regulators should close loopholes that allow advanced AI capabilities to move through foreign subsidiaries.