A Financial Times reports that OpenAI and Google have provided advanced artificial intelligence services to companies tied to Alibaba, Baidu and Tencent through foreign subsidiaries, despite US concerns that the parent companies may have links to the Chinese military. The transactions have highlighted regulatory loopholes that complicate Washington’s efforts to slow China’s technological and military advances in AI.
Both US firms confirmed to the Financial Times that they offer services to subsidiaries of those Chinese technology groups based in places such as Singapore and Hong Kong, while the US government accuses the parent companies of cooperating with the Chinese armed forces.
OpenAI action and suspicion of knowledge distillation
OpenAI told the Financial Times it suspended API access last month for users linked to Alibaba after suspecting misuse, specifically so‑called knowledge distillation — the practice of using outputs from an existing model to train a competing system.
OpenAI said it informed the US government of the matter and reiterated that it does not make its services available within mainland China. It added that it does permit certain China‑owned or China‑headquartered companies to use its tools in third countries where it believes security requirements can be met. The company said it prefers AI shaped by democratic values over software directed by autocratic regimes and argued that citizenship alone should not be an automatic disqualification for access.
Google, Anthropic and the limits of geographic restrictions
Google said its AI services are available in Hong Kong and Singapore provided partners comply with its terms of use, including bans on knowledge distillation. The company acknowledged, however, that geography‑based sales restrictions do not fully prevent abuse because sophisticated actors can circumvent them.
Anthropic has taken a stricter public stance, banning Chinese companies and their foreign subsidiaries from using its advanced models, but it also admits enforcement is challenging. Anthropic previously accused several Chinese AI labs, including DeepSeek, Moonshot and MiniMax, of unauthorized knowledge distillation. In a letter to the US Congress, Anthropic said Alibaba used 25,000 fake user accounts to generate more than 28.8 million message exchanges with the Claude model, a serious breach of its terms of use.
Regulatory gaps and expert criticism
The revelations have revived calls for tighter controls on AI models, analogous to US restrictions on exporting high‑performance chips used for model training. To date, Washington has limited access to certain advanced models — for example, Anthropic’s Mythos and Fable as well as OpenAI’s GPT‑5.6 — but it has not broadly banned the most sophisticated software from Chinese‑linked firms.
The problem is compounded by the so‑called 1260H list compiled by Congress, which names companies tied to the Chinese military yet remains subject to the same enforcement gaps. Chris McGuire of the Council on Foreign Relations argued that cutting‑edge AI models should not be made available to China‑headquartered firms and criticized the Trump administration for emphasizing the need to beat China in AI while failing to take substantive steps to tighten software export controls, which he said would be the most effective tool to slow Beijing.
Why it matters
The Financial Times’ account shows that technically lawful commercial relationships between US tech firms and Chinese‑linked companies can nonetheless be exploited to undermine US efforts to constrain military and advanced AI capabilities. Practices such as knowledge distillation and the use of fake accounts to amass large interaction datasets pose risks for model control and the spread of sensitive technology.
These developments illustrate how current regulatory frameworks struggle to keep pace with evolving AI techniques and complex multinational corporate structures. The debate is likely to continue over stricter access rules, geographic and ownership‑based restrictions, and ways to strengthen enforcement.
Tags: China, artificial intelligence, Alibaba, Tencent, Baidu, Google, OpenAI, Anthropic, knowledge distillation, export controls, 1260H list
Related dates from cited coverage: 2026‑07‑03, 2026‑06‑27, 2026‑06‑25
