OpenAI and Google sell AI access to overseas units of Pentagon-blacklisted Chinese companies

Sam Altman’s OpenAI and Sundar Pichai’s Google have provided powerful AI products to overseas businesses controlled by Chinese corporations named on a US military watchlist.

The customers include Singapore units connected to Alibaba (NYSE: BABA), Baidu (NASDAQ: BIDU), and Tencent Holdings (HKEX: 0700; OTC: TCEHY). Washington claims these three companies have links with China’s armed forces.

OpenAI and Google parent Alphabet (NASDAQ: GOOGL, GOOG) both allegedly confirmed the commercial relationships to the Financial Times.

Nothing about these deals breaks current US law. That is the problem facing lawmakers who want to slow China’s AI growth. Washington limits shipments of the powerful chips needed to build top AI systems. Its rules are far less complete once those systems become online services.

Chinese corporations can still reach American models through businesses registered in approved markets such as Singapore and Hong Kong. The legal opening remains available even when the parent company appears on a Pentagon blacklist.

Washington lets blacklisted Chinese firms reach OpenAI tools through overseas units

The Pentagon utilizes the 1260H list for naming Chinese firms that can provide assistance to the People’s Liberation Army. Although it was required by Congress that the list be maintained, its inclusion does not completely hinder the firm from purchasing advanced software products from America.

The US has put restrictions on specific software products such as the OpenAI GPT-5.6 and Anthropic’s Mythos and Fable. There are no general restrictions on Chinese companies to use the best American AI online.

OpenAI removed access for API users linked to Alibaba last month after spotting activity that may have broken its rules. The company confirmed the suspension this week after the Financial Times reportedly asked about it.

An API simply lets developers use an AI model through the internet. They can send questions, receive responses, and add those results to their own services without downloading the model itself.

The blocked accounts were suspected of carrying out distillation. During that process, a developer gathers responses from one AI system and uses the material to improve another.

OpenAI notified the authorities in the United States about the activity. The company’s models are not available for use within mainland China. It continues to provide its services to some companies that are owned by Chinese citizens in other countries where OpenAI is able to control their activity.

OpenAI stated that its customers “may be active in countries where we can enforce our safeguards and monitor for distillation.” It also pointed out that more people need access to AI systems based on democratic principles rather than those that are used in authoritarian countries.

The company added, “we don’t think nationality alone should decide access.”

Beijing guards Chinese AI methods as American companies adopt cheaper models

Meanwhile, AI models from China’s DeepSeek and Moonshot AI now handle regular work inside Silicon Valley businesses of different sizes, and their lower prices make them useful replacements for services sold by OpenAI and Anthropic. Some companies also run Chinese and American systems together, choosing a different model for each job.

Beijing previously wanted Chinese AI products used across as many countries as possible. Officials treated their international popularity as another way to increase China’s global influence. Many leading Chinese models were released with open code. Developers anywhere could download them for free, inspect how they worked, and use them with few limits.

Chinese officials are now discussing stronger protection for technology created by the country’s leading AI laboratories. They fear that rival governments, criminals, or hostile groups could take valuable methods and later use them against China.

Research papers and public code can reveal technical ideas that took years and large amounts of money to create. Western laboratories have already adopted Chinese methods that help models complete more work with less computing power.

Anthropic has accused DeepSeek, Moonshot AI, and MiniMax of using distillation against its products. In a letter delivered to Congress last month, Anthropic said Alibaba created 25,000 fake accounts that completed more than 28.8 million interactions with Claude. The company said the operation broke its service rules.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.