Blacklisted Chinese Companies Access US AI via Singapore

What happened?

According to an investigation by The Next Web, three Chinese tech giants — included in the US Department of Commerce's Entity List — are purchasing access to advanced AI models from OpenAI (such as GPT-4) and Google (Gemini) through their subsidiaries in Singapore. The operation is carried out via API contracts and software licenses that are not subject to the same restrictions as direct exports from the United States. The Next Web identified that companies like iFlytek, Megvii, and Hikvision — all sanctioned for their ties to the Chinese military — have set up offices in Singapore to channel these purchases. According to documents reviewed by the outlet, an iFlytek subsidiary in Singapore signed an annual contract worth $1.2 million to access GPT-4, while Megvii acquired Gemini licenses for $800,000. These transactions do not directly violate US laws because the subsidiaries are Singaporean legal entities, but they circumvent the spirit of the sanctions.

Why is this important?

This finding exposes a critical vulnerability in US technology export controls. Despite sanctions designed to limit China's access to frontier AI, Chinese companies have found a backdoor in Singapore, a global tech hub with looser regulations. This allows technologies considered strategic for US national security to reach potentially hostile competitors, undermining decades of export control policy. The case is particularly serious because frontier AI, such as GPT-4 and Gemini, has direct military applications: from autonomous reconnaissance systems to large-scale disinformation generation. A 2024 report by the Center for Strategic and International Studies (CSIS) already warned that software export controls were ineffective against intangible transfers. This case confirms it: while hardware like Nvidia chips can be physically tracked, AI models are easily replicated and distributed digitally. Singapore, as a financial and tech hub with lax data transfer oversight, has become the weak link in the sanctions system.

Immediate consequences

• For US companies: OpenAI and Google face legal and reputational risks, as they may be indirectly violating sanctions. Both companies have stated they comply with the law, but the case highlights the difficulty of tracking the end use of their products. If the Department of Justice determines negligence, they could face hefty fines and export restrictions. Moreover, the scandal could accelerate federal AI regulation, such as the proposed 'Export Control of AI Models Act' introduced in Congress in March 2025.
• For Singapore: The country could face diplomatic pressure from Washington to tighten controls, affecting its status as a neutral tech hub. Singapore has traditionally been a center for trade and finance, but its role as an intermediary in sanctions evasion could lead to its inclusion on watchlists. In fact, in 2024, the US Treasury already warned Singapore about using its financial system to evade sanctions on Russia. Now, pressure extends to the tech sector.
• For the global industry: A dangerous precedent is set where export restrictions can be easily bypassed through intermediaries, forcing a review of compliance policies. Companies like Amazon Web Services and Microsoft Azure, which also offer AI APIs, could be dragged into the scandal if their services are found to be used by subsidiaries of sanctioned entities. This could trigger a domino effect of audits and contract cancellations.

Historical context

This is not an isolated case. In 2020, Huawei used Southeast Asian subsidiaries to acquire US chips from companies like Qualcomm, evading sanctions imposed by the Trump administration. The difference now is that AI is an intangible asset, much harder to control than hardware. While chips require physical logistics and can be intercepted at customs, AI models are transferred via lines of code over the internet. The leakage of models like GPT-4 could accelerate China's capabilities in military AI, such as autonomous systems or advanced cyberattacks. In fact, the Pentagon has identified AI as a critical technology for national security, and this case demonstrates that the US technological advantage is eroding. In 2023, China already managed to clone Meta's LLaMA model after its leak, and now it could do the same with GPT-4 or Gemini, narrowing the 2-3 year gap that currently separates the two countries in AI.

What should readers know?

Users and developers consuming APIs from OpenAI or Google Cloud should be aware that their data may be processed on servers serving sanctioned entities, posing security and compliance risks. For example, if a US company uses OpenAI's API and its data passes through a node in Singapore that also serves iFlytek, there could be risks of sensitive commercial data leakage. Additionally, this case underscores the need for stricter regulation on AI software distribution, not just hardware. The Federal Trade Commission (FTC) has already launched a preliminary investigation into OpenAI for potential sanctions violations. For investors, shares of Alphabet and Microsoft could be affected if fines or restrictions are imposed. On the geopolitical front, this case could lead to escalating tensions between the US and China, with trade retaliation or new tech sanctions.

"AI knows no borders, but sanctions do. Singapore has become the digital Trojan horse of US technology into China." — Analyst at TheVortiq

In summary, this case exposes a systemic fragility in AI export controls. While attention focused on hardware, software transfers have remained unprotected. The solution will require closer international cooperation, harmonized regulations, and perhaps the creation of a 'global licensing regime' for frontier AI. Until then, sanctions evasion will remain a cat-and-mouse game, with Singapore at its epicenter.