China’s approach to governing artificial intelligence has always been notable for its speed. While Western regulators have spent years debating frameworks, Beijing has moved to put rules on the books — and this week it took another step forward, publishing draft regulations specifically designed for interactive AI services, the category that encompasses AI chatbots, virtual assistants, and any product built around real-time conversational AI.
The draft rules, published by the Cyberspace Administration of China (CAC) and open for public comment, extend the Generative AI Regulations, which came into force in August 2023. Those earlier rules established baseline obligations for providers of generative AI services in China, covering issues such as content safety, algorithmic transparency, and data handling. The new draft goes further, targeting the specific characteristics of interactive services — products where users engage in ongoing, real-time dialogue with an AI system.
According to an analysis published by the law firm Mayer Brown, the draft imposes several notable requirements on service providers. Mandatory content filtering systems must be in place to prevent the generation of prohibited content, with particular emphasis on material that could be considered politically sensitive, socially destabilizing, or in violation of China’s existing internet content rules. Users must be clearly informed that they are interacting with an AI system rather than a human, a requirement that has parallels in emerging regulations in the European Union and elsewhere.
Identity verification requirements are also included, consistent with China’s broader policy of linking online activity to verified real-world identities. Providers will be required to maintain logs of user interactions and to make these available to regulators upon request, a provision that has drawn attention from privacy advocates and international businesses operating in China.
The draft also includes provisions around the training data used to develop interactive AI systems, requiring that data collection and use comply with existing personal information protection laws. This is significant given the scale of data large language models require and the ongoing global debates over the legality of scraping web content for AI training.
For domestic AI companies — including Baidu, whose ERNIE Bot is one of China’s most widely used AI chatbots, as well as newer entrants like Moonshot AI’s Kimi and Zhipu AI’s GLM series — the new rules represent an additional compliance layer. Most large providers already operate under the 2023 Generative AI Regulations and have established content moderation systems; the new draft largely formalizes and extends obligations they are already working to meet.
The more significant implications may be for smaller developers and for international companies seeking to offer interactive AI services in China. The compliance burden — particularly around identity verification, log retention, and government data access — is substantial, and the requirements around content filtering are difficult to meet without deep familiarity with China’s evolving content governance landscape.
China’s AI regulatory approach has attracted growing international attention. The country has moved faster than most to put binding rules in place, and its framework, which combines detailed technical requirements with strong state oversight, represents a distinct model compared to the risk-based, principles-oriented approach favored by the EU or the largely voluntary framework that has characterized US policy to date. As interactive AI services become ubiquitous globally, the question of which regulatory model prevails, or whether some form of convergence is possible, will be one of the decade’s defining governance questions.
The public comment period for the draft rules closes in early May. Final regulations are expected to be published later in 2026.