A Financial Times investigation published July 10, 2026, has revealed that OpenAI and Google have continued providing advanced AI model access to Singapore-based subsidiaries of Chinese tech companies that the Pentagon has placed on its blacklist, exposing a significant gap in U.S. export control enforcement.

The report found that a subsidiary incorporated in Singapore can sign contracts and pay API bills in ways its mainland China-based parent company cannot, effectively bypassing restrictions designed to keep advanced American AI technology out of the hands of entities deemed a national security risk. The loophole exists because U.S. export restrictions on AI have been built around specific entities, specific geographic locations, and physical exports — not software services delivered through cloud APIs. For the latest developments on this story and broader AI industry coverage, the implications stretch from Silicon Valley to Washington.

How the Singapore Loophole Works

Under current U.S. rules, mainland China faces restrictions on accessing frontier AI models. Singapore does not. The Pentagon's so-called 1260H blacklist restricts U.S. defense contracting with listed companies and adds reputational pressure, but according to reporting by the Associated Press, it does not automatically prevent those companies from conducting ordinary commercial business through overseas affiliates.

OpenAI told the Financial Times that it blocks direct access to its models from China but allows some Chinese-owned companies to use its services in jurisdictions where it says safeguards can be enforced and misuse can be monitored. The company recently cut off API access for users linked to Alibaba after identifying suspected distillation — the practice of querying a rival model to collect its outputs and train a competing system — and reported the activity to the U.S. government.

Google's position proved looser. The company said its AI services remain available in markets including Singapore and Hong Kong, subject to rules that prohibit distillation. However, Google also acknowledged that location-based restrictions can be bypassed, an admission that arguably matters more than the policy language itself.

Anthropic Takes the Opposite Approach

Anthropic has taken a starkly different stance, barring Chinese-owned entities outright — including foreign subsidiaries. There is no Singapore workaround for Claude. The company hardened this position after a specific dispute with Alibaba's Qwen lab.

According to Business Insider, Anthropic's June 10 letter to U.S. Senators Tim Scott and Elizabeth Warren alleged that Alibaba-affiliated operators used nearly 25,000 fraudulent accounts to generate 28.8 million illicit interactions with Claude between April and June 2026. Anthropic said the alleged campaign was designed to extract Claude's capabilities for Alibaba's own models. Alibaba has disputed U.S. military-link claims in other contexts.

The Export Control Gap

The disclosures have reopened a debate in Washington over whether export controls should cover AI software and services the way they already cover physical hardware such as lithography equipment and GPUs. Current rules were designed for a hardware world: tangible components crossing a border in a shipping container. A Singapore subsidiary buying API credits does not fit that model.

Lawmakers and policy experts argue that if a Singapore unit of a Pentagon-listed Chinese company can access tools its mainland parent cannot, the regulatory framework is lagging behind the market. For now, the loophole holds: a subsidiary registered in Singapore can still purchase what its parent in Shenzhen or Hangzhou cannot, and nothing in current U.S. law appears to explicitly prohibit it.

What Happens Next

The FT report has intensified calls for Congress to close the gap between hardware-focused export controls and the reality of cloud-delivered AI services. Several members of Congress have previously introduced legislation aimed at extending export restrictions to cover AI software and API access, though no such bill has yet reached a floor vote.

OpenAI and Google maintain they are following current law while monitoring for misuse. Anthropic's position — a blanket prohibition on Chinese-owned entities — has drawn praise from national security hawks but criticism from those who argue it sacrifices oversight by cutting off access entirely. The debate underscores a fundamental tension in U.S. AI policy: whether monitored engagement or total separation is the more effective strategy for managing technology flows to geopolitical rivals.

The Singapore loophole may be legal, but as the FT investigation makes clear, it is a vulnerability that Washington is increasingly unwilling to tolerate.

---

Stay Ahead of AI — For continuous coverage of AI policy, export controls, and the global technology race, visit AI Buzz Wire. Read more AI news →