China’s top state security authority warns of privacy and data risks from ‘AI relay platforms’
China’s top state security authority warns of privacy and data risks from ‘AI relay platforms’
https://www.globaltimes.cn/page/202606/1363006.shtml
Publish Date: 2026-06-07 21:09:00
Source Domain: www.globaltimes.cn
Photo: WeChat account from Ministry of State Security
China’s Ministry of State Security (MSS) warned on Monday that the rapidly expanding market for “AI relay platforms” currently faces risks that should not be overlooked, including inadequate operating qualifications, weak security protections, frequent privacy leaks and the resale of user data.
An “AI relay platform” acts as an intermediary layer between users and official services provided by AI model developers, according to MSS. As the demand for AI applications surged in recent months, “AI relay platforms” that provide bulk access to both domestic and overseas large language models have rapidly gained popularity in China.
These platforms integrate application programming interfaces from multiple AI model providers into a single platform and offer them to users. An “AI relay platform” functions as an intermediary between users and large models, allowing users to access multiple models more conveniently and meet diverse needs through a one-stop service.
Their advantages include eliminating the need to switch between models, offering relatively low prices, supporting mainstream domestic payment methods and bypassing certain usage restrictions. However, their rapid and loosely regulated expansion has also created a range of security risks.
The MSS warned that such platforms could expose user data and lead to privacy breaches. As third-party gateways, “AI relay platforms” store user-submitted data on their own servers. Some lack proper encryption and data governance mechanisms, while others may retain user data without authorization and resell it to other model developers for training purposes, increasing the risk of privacy leaks.
Some operators, seeking to reduce costs and maximize profits, substitute lower-end models for premium ones, cut computing resources or disable verification functions. These practices may lead to less accurate outputs, weaker logical consistency and could…
