In a revelation that underscores the complexities of enforcing global technology restrictions, major US artificial intelligence developers OpenAI and Google have confirmed that they have provided advanced AI services to Singapore-registered subsidiaries of prominent Chinese technology conglomerates, as first reported by the Financial Times. The companies involved (Alibaba, Baidu, and Tencent) are all currently listed on the Pentagon’s 1260H blacklist, a designation reserved for entities with alleged ties to China’s military apparatus.
The Singapore Loophole: How Blacklisted Firms Access US AI
This development highlights a significant legal loophole in current United States export controls. While direct access to frontier AI models from mainland China is heavily restricted, the regulations do not explicitly prohibit Chinese-owned companies from purchasing these services through their international subsidiaries, provided those subsidiaries are located in jurisdictions that are not restricted.
The situation has intensified the ongoing debate surrounding the efficacy of US efforts to curb Beijing’s access to cutting-edge artificial intelligence technology, a sector increasingly viewed as critical to national security and economic competitiveness.
Navigating the Regulatory Gray Area
The core of the issue lies in the structure of US export controls, which are primarily geographically based rather than entity-based when it comes to software and services. While the US government has implemented stringent measures to block direct access to frontier models—such as OpenAI’s GPT-5.6 and Anthropic’s Mythos and Fable—from within mainland China, these rules do not automatically extend to foreign subsidiaries of Chinese firms.
This regulatory gray area allows companies like Alibaba, Baidu, and Tencent to legally access advanced AI capabilities through their Singaporean outposts. The practice is entirely lawful under current regulations, but it has sparked renewed calls from national security experts and policymakers for tighter, more comprehensive export controls on advanced AI models.
The recent news that Pentagon adds Alibaba, Baidu and BYD to military company List in sweeping escalation has only heightened scrutiny on these transactions. Critics argue that allowing blacklisted entities to access top-tier AI, regardless of the subsidiary’s geographic location, undermines the fundamental purpose of the restrictions.
OpenAI’s Response and the Threat of “Distillation”
OpenAI, the creator of ChatGPT, has acknowledged the situation and detailed its approach to managing access. The company stated that while it prohibits access to its models from within China, it permits “some companies” with Chinese ownership to use its tools in countries where it can “enforce safeguards and monitor for distillation.”
“Distillation” is a critical concern in the AI industry. It refers to the practice where developers use the output generated by a highly advanced AI model to train and improve their own competing systems. This essentially allows a company to piggyback on the massive research and development investments made by frontier model developers.
OpenAI revealed that it suspended API access for Alibaba-affiliated users last month after detecting suspected distillation activities. The company reported these suspicions to the US government, emphasizing its preference for AI shaped by democratic values rather than autocratic control, while maintaining that nationality alone should not dictate access.
Diverging Approaches Among US AI Leaders
The response to this regulatory challenge varies significantly among leading US AI developers. While OpenAI and Google navigate the complexities of monitoring usage in jurisdictions without restrictions, Anthropic has adopted a notably stricter stance.
Anthropic, the developer behind the Claude series of models, has implemented a blanket ban on Chinese companies and any foreign entities they own from accessing its frontier models. This aggressive approach follows the company’s description of attempts by Chinese firms to bypass its safeguards. Anthropic has previously accused several Chinese AI companies of distillation, highlighting the persistent threat of intellectual property extraction in the global AI race.
Google, for its part, confirmed that its AI services are available in locations like Hong Kong and Singapore, subject to internal policies prohibiting distillation and misuse. However, the company acknowledged the limitations of geographic restrictions, noting that sophisticated users can often find ways to circumvent them.
The Push for Tighter Export Controls
The disclosures regarding OpenAI and Google’s dealings with subsidiaries of blacklisted Chinese firms have amplified calls for the US administration to implement broader export controls on advanced AI software. Proponents of stricter regulations argue that the current framework is insufficient to prevent Chinese companies from extracting advanced AI capabilities without bearing the associated development costs.
Experts warn that if left unchecked, these practices could erode the economic foundation of US leadership in frontier AI. The challenge for policymakers is to craft regulations that effectively restrict access to sensitive technology by adversarial entities without stifling global innovation or unduly penalizing legitimate international business operations.
As the AI landscape continues to evolve at a breakneck pace, the tension between open technological development and national security imperatives will only intensify. The “Singapore loophole” serves as a stark reminder of the difficulties inherent in regulating a technology that transcends traditional geographic boundaries.