cd /news/ai-policy/openai-and-google-sold-ai-access-to-… · home topics ai-policy article
[ARTICLE · art-54249] src=startupfortune.com ↗ pub= topic=ai-policy verified=true sentiment=↓ negative

OpenAI and Google Sold AI Access to Blacklisted Chinese Firms via Singapore

OpenAI and Google have been providing advanced AI model access to Singapore-registered subsidiaries of Chinese firms Alibaba, Baidu, and Tencent, which are on the Pentagon's list of companies with ties to China's military. The arrangement exploits a gap in US export controls that restrict AI access to mainland China but not Singapore, allowing these subsidiaries to use services their parent companies cannot. OpenAI has cut off one such account after detecting model distillation, while Google has not disclosed any enforcement actions.

read3 min views1 publishedJul 10, 2026
OpenAI and Google Sold AI Access to Blacklisted Chinese Firms via Singapore
Image: Startupfortune (auto-discovered)

OpenAI and Google have been supplying advanced AI model access to Singapore-registered units of Alibaba, Baidu and Tencent, three Chinese firms the Pentagon says have ties to China's military.

Here's the loophole nobody closed. Alibaba, Baidu and Tencent all sit on the Pentagon's 1260H list, the roster of Chinese companies the US government says have links to the People's Liberation Army. Their Shenzhen and Hangzhou headquarters can't touch OpenAI's or Google's most advanced models. Their Singapore subsidiaries can. According to a Financial Times report published Friday, that gap has already been used, and it isn't illegal.

The mechanism doesn't require a law degree to follow. A subsidiary incorporated in Singapore operates under Singapore law, not Chinese law. It can sign a contract, open an API account and pay a bill in a way its parent company on the mainland cannot. US export controls on AI target specific entities and specific geographies. Mainland China is restricted. Singapore is not. Nothing in the current rules stops a Singapore-registered arm of Alibaba from doing what Alibaba itself is barred from doing.

OpenAI told the Financial Times it blocks direct access to its models from China but allows some Chinese-owned firms to use its services in jurisdictions where it says it can enforce safeguards and monitor for misuse. That policy already produced one enforcement action. Last month OpenAI cut off API access for users linked to Alibaba after identifying suspected distillation, the practice of querying a rival's model to harvest its outputs and train a competing system on the results. OpenAI reported the activity to the US government.

Google's answer is looser. The company said its AI services remain available in markets including Singapore and Hong Kong, subject to usage policies that also prohibit distillation. Unlike OpenAI, Google hasn't disclosed any enforcement action against a specific Chinese-linked account. TipRanks reported Alphabet shares slipped in premarket trading Friday as the report spread, a sign investors are pricing this as more than a compliance footnote.

Anthropic took the opposite approach.

Anthropic bars Chinese-owned entities outright, foreign subsidiaries included. No Singapore workaround. The policy hardened after a specific incident. In a June letter to Capitol Hill, Anthropic accused Alibaba's Qwen lab of running what it called the largest known distillation campaign against its Claude models to date, alleging that roughly 25,000 fraudulent accounts generated more than 28.8 million exchanges with Claude, according to CNBC's reporting on the letter. Alibaba has since blocked Claude Code for its own developers, leaving both companies walled off from each other's systems.

The difference between the labs isn't caution versus recklessness. It's a bet about where the real risk sits. OpenAI and Google are wagering that a monitored, revenue-generating relationship with a Singapore entity is safer than losing visibility into how Chinese firms reach frontier models altogether. Anthropic is wagering that any relationship at all is a door that gets forced open wider than intended. Frankly, the Alibaba letter suggests Anthropic's bet has more evidence behind it right now.

The disclosures have reopened a fight in Washington over whether export controls should cover AI software and services the way they already cover chips. Current rules were built for hardware, for lithography equipment and GPUs crossing a border in a shipping container. A Singapore subsidiary buying API credits doesn't fit that model at all, and lawmakers pushing for tighter China policy have signaled the Singapore structure is exactly the kind of gap they want closed next.

For now, the loophole holds. A subsidiary registered in Singapore can still buy what its parent in Shenzhen or Hangzhou cannot, and nothing in current US law says otherwise. Also read: US venture capital hit a record in 2026, and almost none of it trickled downFounders Have Enough Dashboards, What They Want Now Is a DiagnosisRobbyant Builds Out an Embodied-Native Full-Stack for Physical AI

── more in #ai-policy 4 stories · sorted by recency
── more on @openai 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/openai-and-google-so…] indexed:0 read:3min 2026-07-10 ·