Draft Chinese language AI Guidelines Define ‘Core Socialist Values’ for AI Human Persona Simulators

As first reported by Bloomberg, China’s Central Our on-line world Affairs Fee issued a doc Saturday that outlines proposed rules for anthropomorphic AI systems. The proposal features a solicitation of feedback from the general public by January 25, 2026.

The principles are written usually phrases, not legalese. They’re clearly meant to embody chatbots, although that’s not a time period the doc makes use of, and the doc additionally appears extra expansive in its scope than simply guidelines for chatbots. It covers behaviors and general values for AI merchandise that interact with individuals emotionally utilizing simulations of human personalities delivered by way of “textual content, picture, audio, or video.”

The merchandise in query needs to be aligned with “core socialist values,” the doc says.

Gizmodo translated the doc to English with Google Gemini. Gemini and Bloomberg each translated the phrase “社会主义核心价值观” as “core socialist values.”

Beneath these guidelines, such methods must clearly establish themselves as AI, and customers should be capable to delete their historical past. Individuals’s information wouldn’t be used to coach fashions with out consent.

The doc proposes prohibiting AI personalities from:

• Endangering nationwide safety, spreading rumors, and inciting what it calls “unlawful spiritual actions.”
• Spreading obscenity, violence, or crime
• Producing libel and insults
• False guarantees or materials that damages relationships
• Encouraging self hurt and suicide
• Emotional manipulation that convinces individuals to make unhealthy choices
• And Soliciting delicate data

Suppliers wouldn’t be allowed to make deliberately addictive chatbots, or methods meant to exchange human relationships. Elsewhere, the proposed guidelines say there should be a pop-up on the two hour mark reminding customers to take a break within the occasion of marathon utilization.

These merchandise additionally must be designed to choose up on intense emotional states and hand the dialog over to a human if the person threatens self-harm or suicide.