OpenAI and Google Gave AI Models to Blacklisted Chinese Tech Giants via Singapore

The Financial Times published an investigation on July 10, 2026, revealing that OpenAI and Google had been providing advanced artificial intelligence services to 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 names entities allegedly linked to the Chinese military.

The mechanism uncovered by FT reporters relies on a legal loophole that opened up after the Trump administration rolled back the so-called diffusion rule earlier this year, a rule introduced under Joe Biden. That rule was specifically meant to limit this kind of export-control workaround through foreign subsidiaries.

How the Singapore Loophole Works

On paper, a Chinese conglomerate's Singapore subsidiary is a fully Singaporean entity, operating under Singaporean law, paying Singaporean taxes, and able to sign contracts unavailable to its parent company in Shenzhen or Hangzhou. It's this legal structure that allowed Alibaba, Baidu and Tencent to legally access OpenAI and Google's most advanced models, even though direct access from mainland China remains blocked.

Current US export controls focus on specific entities and locations rather than the technology itself, which in practice means transactions through foreign subsidiaries don't break the law, even when the parent company sits on a military blacklist.

Company Responses

OpenAI told the Financial Times it blocks direct access to its models from Chinese territory but allows some companies controlled by Chinese entities to use its services in jurisdictions where safeguards can be enforced. The company also said it suspended access for an Alibaba-linked firm after detecting attempts at so-called distillation, using responses from advanced models to train competing systems, and reported the matter to US authorities.

Geographic restrictions alone aren't enough to stop sophisticated users from working around controls - Google spokesperson

Google confirmed its AI services are available in markets such as Singapore and Hong Kong, while stressing compliance with its usage policies. Anthropic, a competitor to both companies, has taken a considerably more restrictive approach, barring Chinese firms and the foreign entities they control from accessing its most advanced models, while also calling for broader export controls in this area.

Policy Debate in Washington

The FT's findings have reignited debate in Washington over imposing controls on access to advanced AI models similar to those that have long governed trade in advanced semiconductors. Critics of the current rules argue the loophole lets Chinese firms with military ties indirectly benefit from the most advanced American AI technology, despite formally sitting on the Pentagon's sanctions list.

The market's reaction to the investigation was muted. Alphabet shares were nearly flat in pre-market trading, as were Alibaba's US-listed shares, suggesting investors aren't expecting swift, hard-hitting regulation in this area.

The case fits a broader pattern in which Chinese tech firms have been moving the training of advanced language models to Southeast Asia, mainly Singapore and Malaysia, to legally access Nvidia's most powerful chips, which aren't directly available in mainland China. Earlier reporting had already shown Alibaba and ByteDance pursuing a similar strategy for computing hardware, and the FT's latest investigation shows the same geographic loophole mechanism now applies to access to the AI models themselves, not just the chips.

For European and Polish companies using OpenAI and Google models, the case carries only indirect relevance, but it shows that global rules on AI access still have significant gaps, and closing them in practice falls to the policies of individual model providers rather than any unified international law.

Sources: Investing.com, dev.ua, Il Sole 24 Ore (en.ilsole24ore.com)