China warns developers over overseas AI relay services

China’s top security authority has warned developers about risks linked to AI relay services that provide access to overseas models. The Ministry of State Security said some platforms could expose user information and enable unauthorized cross-border data transfers. 

• China’s Ministry of State Security warned developers about risks from overseas AI relay services.
• The ministry said weak security controls could expose user data and enable illegal data trading.
• Chinese developers still seek U.S. AI models for coding despite advances by local AI firms.

The warning comes as Chinese developers continue seeking access to foreign AI systems despite domestic competition. Authorities said some operators lack proper safeguards and could expose sensitive information.

Ministry of State Security raises data security concerns

According to a notice published on the Ministry of State Security’s official WeChat account, relay services act as intermediaries between developers and AI providers. The ministry said these platforms combine access to several domestic and foreign models through one interface. While the services offer convenience and lower costs, officials said some operators present security concerns. The ministry stated that weak controls could increase privacy risks and facilitate illegal data trading. It also urged developers to assess platform credentials before sharing information.

The ministry said some operators store user information on servers without sufficient encryption. As a result, unauthorized parties could access or misuse sensitive data. Officials also warned that some services may transfer information across borders without proper authorization. The notice described these practices as potential threats to data security. Authorities encouraged developers to strengthen protections for confidential information.

Another concern involves model substitution by relay providers. The ministry said some operators may replace premium models with cheaper alternatives. Such practices could mislead customers about the services they receive. Officials stated that users may experience reduced performance without knowing a replacement occurred. The ministry included this issue among several operational risks facing developers.

Developers continue seeking access to US models

Chinese developers increasingly use relay services to access foreign AI systems unavailable in mainland China. Anthropic’s Claude and OpenAI’s GPT models do not officially support users in the country. Despite local progress, developers continue choosing foreign models for coding tasks. Industry participants frequently cite performance differences in software development workloads. This demand has supported growth in the relay services market.

Recent rankings from Artificial Analysis highlighted continued interest in U.S. models. The benchmark platform placed products from OpenAI, Anthropic, and Google at the top of its Coding Index. GPT-5.5, GPT-5.4 Mini, Claude Opus 4.8, Claude Opus 4.7, and Gemini 3.1 Pro ranked highly. The results showed continued competition between American and Chinese developers. Chinese companies also maintained positions within the rankings.

Among domestic systems, Alibaba’s Qwen3.7 Max achieved the highest placement. DeepSeek’s V4 Pro followed closely behind in the rankings. Chinese developers continue evaluating both local and foreign options for programming tasks. Performance remains an important factor when selecting AI systems. Demand for coding tools has therefore remained active across the market.

White House and Anthropic cite access attempts

The ministry also warned about hidden backdoors within some relay platforms. Officials said these features could bypass authentication controls and compromise user accounts. Such tools may also allow surveillance of developer devices and activity. Authorities advised developers to use trusted providers with stronger security measures. The notice stressed the importance of protecting personal and commercial information.

Outside China, U.S. officials have also addressed access concerns involving advanced AI systems. The White House said in April that Chinese entities sought access through networks of intermediary accounts. According to the administration, those efforts targeted advanced American AI technologies. Officials said the activity involved coordinated methods designed to bypass restrictions. The warning highlighted ongoing scrutiny around access controls.

Anthropic has previously reported similar concerns regarding access attempts. The company disclosed efforts by China-linked actors to reach its systems through proxy infrastructure. Those attempts involved coordinated networks designed to obtain model access. The disclosure came as developers increasingly searched for alternative pathways to foreign AI products. Both the White House and Anthropic have continued monitoring such activity.