SINGAPORE (Realist English). US tech corporations OpenAI and Google continued to supply advanced artificial intelligence models to subsidiaries of Alibaba, Baidu and Tencent, even though the parent companies of these Chinese giants are on the Pentagon’s so-called “1260H list” — a roster of enterprises Washington considers linked to China’s armed forces.
According to the Financial Times, citing confirmation from the companies themselves, the clients in question are Singapore-registered subsidiaries of the Chinese corporations.
The Singapore Loophole: How AI Bypasses Sanctions
Supplies were carried out via Singapore, which has become a kind of “neutral zone” in the AI arms race. US export restrictions target specific organisations and geographic regions — mainland China is restricted, but Singapore is not subject to the bans.
All three Chinese tech giants have significant operations in the city‑state.
The legal distinction is crucial: the Singaporean subsidiary of a blacklisted Chinese firm is, on paper, a Singaporean company. It operates under local laws, pays local taxes and can enter into contracts that are unavailable to its parent company in Shenzhen or Hangzhou.
Such sales remain legal because current US export control rules do not yet prohibit access to advanced AI software, even for companies on the “1260H list.”
However, this case has exposed a serious gap in the US export control system and has intensified calls to extend restrictions — similar to those already in place for advanced chips — to AI models.
OpenAI and Google’s Position: Access Granted, with Conditions
Following the FT’s inquiry, OpenAI said it had blocked access to its API for Alibaba‑affiliated users last month over suspicions of “distillation” — using AI model responses to train competing systems. The company reported this activity to the US government.
OpenAI stressed that it does not provide direct access to its models in China, but allows individual companies with Chinese owners to use its services in countries where it can monitor compliance and track distillation attempts. The company also said it does not consider nationality the sole criterion for restricting access to its models.
Google said its AI services are available in Hong Kong and Singapore in accordance with usage rules prohibiting distillation. At the same time, the company acknowledged that geographical restrictions alone are insufficient, as they can be circumvented.
Unlike OpenAI and Google, Anthropic has banned Chinese companies and their overseas subsidiaries from using its advanced models. However, even Anthropic acknowledged that enforcing this ban is difficult and last week began closing loopholes that allowed its restrictions to be bypassed.
The Microsoft Precedent: Parallel Supply Channels
Microsoft has operated under similar conditions for several years, maintaining partnerships that allow it to offer OpenAI‑based models directly in China, despite restrictions preventing OpenAI from operating there directly. Microsoft’s Azure cloud serves as the delivery mechanism. Since Microsoft holds exclusive commercial licensing rights to OpenAI’s models, it can distribute them through its existing operations in China.
Tencent Eyes Manus: A $2 Billion Buyback
Alongside the AI model supplies, another drama is unfolding. Tencent is in talks to become the largest shareholder in Manus — a Chinese AI startup in the agentic AI space whose acquisition by Meta was blocked by Beijing in April.
Most of Manus’s early investors, including Tencent, the venture capital firm ZhenFund and HSG (formerly Sequoia Capital China), are discussing a deal that would unwind Meta’s purchase at the same $2 billion valuation.
According to the Financial Times, citing two sources familiar with the talks, Tencent would acquire the largest stake but would remain a minority shareholder. Manus would continue to operate independently from Singapore and would not be integrated into Tencent’s business.
Investors supporting the deal expect Manus to continue independent development and eventually hold an IPO in Hong Kong. According to sources, Manus’s annual recurring revenue is approaching $500 million, reflecting significant growth since Meta first attempted to acquire the company.
Beijing Tightens Control
The Chinese authorities’ decision to block Meta’s Manus acquisition is part of a broader trend of tightening control over domestic AI assets. Chinese officials have held meetings with leading tech companies on possible restrictions on foreign access to the country’s most advanced AI models.
In April, China’s state planner ordered Meta to unwind its completed acquisition of Manus, initiating a complex process to reverse the deal. Chinese authorities also launched investigations into Manus and other local AI startups that had moved overseas to determine whether they had violated export control laws.
The two stories that unfolded on July 10, 2026, highlight the paradoxes of the global AI race. On one hand, US companies are using legal loopholes to supply advanced AI models to Chinese tech giants on the “blacklist.”
On the other, Beijing is blocking the acquisition of Chinese AI assets by US corporations, seeking to maintain control over strategic technologies. Singapore is becoming a neutral zone where the interests of the two superpowers converge, and legal formalities are becoming the main battleground in the technology war.