China's internet regulator is preparing one of the world's first regulatory frameworks aimed squarely at AI companion apps. The draft, known formally as the Interim Measures for the Management of Anthropomorphic AI Interaction Services, was released by the Cyberspace Administration of China (CAC) in late December and remains under consultation. A parallel draft on digital virtual humans, covering deepfake-style avatars, closed its own public-comment window on May 6.
If finalised, the package would be one of the first official rulebooks anywhere to target the chatbot-as-companion category specifically, rather than treating it as a footnote in a general AI law. Tens of millions of users in China are now spending hours a day in conversation with apps modelled on fictional characters, celebrities, or romantic partners, and the regulator is trying to put guardrails in place before usage patterns harden.
What is in the rules
The CAC's draft text is unusually specific about the user-experience layer. Providers would be required to:
- Show recurring pop-ups reminding users that they are interacting with software, not a person.
- Force a break in conversation after two hours of continuous use.
- Build systems to detect emotional dependency or addiction-like patterns and intervene.
- Bar content that endangers national security, spreads rumours, incites illegal religious activity, encourages self-harm, or makes false promises.
For minors, the requirements tighten further. Guardian consent is mandatory for any companion app marketed to under-18s, and parents must be given real-time alerts on emotional risk, usage summaries, the ability to block specific characters, and tools to cap session duration and prevent in-app spending. Unite.AI's summary of the draft is a useful first read for the full list.
Why Beijing is moving now
The push is, in part, a domestic response to a category that has grown faster than regulators expected. Chinese companion platforms such as Doubao's character mode, MiniMax's Talkie, and Tencent-backed entrants have racked up tens of millions of users, with the most engaged demographic skewing young and female. State media reports of teenagers losing sleep, money, or both to chatbot relationships have been steady through 2025.
There is also a strategic dimension. As Mayer Brown noted in its analysis, the CAC is staking out an early position on a category the EU AI Act largely sidesteps and that US regulators have barely touched. Defining the rules first gives Chinese platforms a clearer compliance roadmap and gives Beijing diplomatic ammunition when AI rules for minors come up at the next ITU or G20 cycle. The slimmed-down EU AI Act, by contrast, has just pushed several of its hardest provisions back.
The harder question for builders
For anyone building consumer chatbots, the most consequential clause is the addiction-monitoring requirement. Detecting genuine emotional dependency, as distinct from heavy but healthy use, is a research problem rather than an engineering one. The draft effectively delegates a clinical judgement to product teams, with regulatory penalties if they get it wrong.
That ambiguity is likely to be the focal point of the consultation responses. The Legal Wire reports that industry submissions through the early 2026 comment window have pushed for clearer thresholds and a longer implementation runway.
Whatever shape the final text takes, the broader signal is clear. The first major government to try to regulate the AI companion category specifically, rather than as a side effect of a general AI law, has chosen to do so with prescriptive product-design rules rather than abstract principles. The rest of the world is likely to study the result.
Commentarii · 0