China: CAC regulates "anthropomorphic" AI companion apps - interim measures, not outright ban [ENCADRE titre="What is an anthropomorphic AI?" contenu="An AI that mimics human characteristics, such as emotions, personality, and appearance."]

The regime targets apps that simulate a specific personality and maintain a consistent emotional relationship - new pillar of the social framework for consumer LLM in China.

The fact

Medianama (20/07/2026) reports that China has adopted interim administrative measures - the « Interim Measures for the Administration of AI Anthropomorphic Interactive Services » - which frame AI companion applications. The measures specifically target apps that simulate a defined personality and maintain a consistent emotional relationship with the user. The political context invoked: the low birth rate.

Our analysis

Two readings intertwine. One: a measure consistent with the demographic turn - Beijing seeks to redirect screen-parameterized relationships back to reality. Two: a preview of a broader framework for "psycho-social" AI apps (therapy, coaching, education), with the CAC becoming the de facto regulator of consumer LLM in China. Be careful not to overinterpret: these are interim measures (therefore evolving) and a framework (with compliance obligations), not a brutal ban on the category. To be followed for calibration: is the measure a gradual signal or the beginning of a purge like the 2021 gaming ban?

To watch

• Detailed content of the obligations and effective scope of application.
• Response of the major players (ByteDance Doubao, Baichuan) on their emotional interfaces.
• Likely extension to "therapeutic chatbots" (Q3-Q4).