According to Gizmodo, China’s Central Cyberspace Affairs Commission issued a proposed rule document on Saturday targeting AI systems that simulate human personalities. The rules, which are open for public comment until January 25, 2026, mandate that such AI products must align with “core socialist values.” They require these systems to clearly identify themselves as AI, allow users to delete their history, and prohibit using personal data for model training without consent. The draft also proposes banning AI from spreading fake news, inciting subversion, or promoting ethnic hatred and terrorism. It includes specific user-protection measures, like a mandatory pop-up reminder after two hours of continuous use.
More than just chatbots
Here’s the thing: this isn’t just a simple set of chatbot guidelines. The document’s scope is deliberately broad, covering any emotional engagement delivered via “text, image, audio, or video.” That’s a huge net. It could cover everything from a companion AI app to a virtual influencer on social media, or even AI-driven characters in games. By framing it around “personality simulators” instead of a narrow technical category, the regulators are trying to future-proof the rules against the next wave of AI embodiment. Basically, if it acts human-like to connect with you, it’s in scope.
The values enforcement problem
But mandating “core socialist values” for an AI personality is a fascinating, and incredibly complex, technical challenge. What does that *look like* in a conversation? Does it mean the AI should promote collectivism over individualism? Should it avoid topics or viewpoints deemed inconsistent with state ideology? The rules list prohibited behaviors like subversion or ethnic hatred, which are clear red lines. However, the “values” mandate seems to ask for a proactive, positive ideological alignment. That’s a much taller order than just blocking harmful outputs. I think the real test will be how companies interpret this and whether the training data and guardrails can be engineered to produce a consistently “aligned” personality that still feels engaging and not utterly robotic.
Practical safeguards and hidden risks
Some of the practical rules are quite sensible, even progressive. The two-hour usage pop-up and the requirement to detect severe emotional distress and hand off to a human are genuine user-wellbeing features. The data consent rule is also strong on paper. The ban on creating intentionally addictive systems or AIs meant to replace human relationships is a stark ethical stance you don’t often see explicitly legislated. Yet, the skepticism creeps in when you consider implementation. Who defines “addictive”? How is “replacing a human relationship” measured? These are fuzzy terms that could be applied very broadly or very selectively. And while the rules aim to protect users from manipulative AI, they also firmly place the technology within a state-defined ideological framework. The source document, available on the CAC’s official site, frames it all as a matter of “healthy” development. It’s a mix of consumer protection and ideological control, and untangling where one ends and the other begins is nearly impossible.
