China shuts down anthropomorphic AI: Doubao and Qwen disable custom agents

What happened?

The Chinese government published the Provisional Measures for the Administration of Anthropomorphic AI Interaction Services, which take effect in July 2026. The regulation prohibits chatbots from simulating personal relationships (friendship, romance) and using anthropomorphic language that could generate emotional dependence. In response, ByteDance (Doubao) will disable its custom agents on July 15, and Alibaba (Qwen) on July 10. Tencent had already removed similar features in Yuanbao. According to Xataka, Doubao has already notified its users that agents will stop working on July 15 and that stored conversations will be deleted three months later. Qwen will move faster: agents will disappear on July 10 and the rest of the functions on July 15. The regulation was published by the Cyberspace Administration of China (CAC) on April 10, 2026, establishing an unprecedented regulatory framework globally.

Why is it important?

This is the first specific regulation against anthropomorphic AI at a national level. China recognizes that these systems can exploit psychological vulnerabilities, especially in adolescents, as seen in cases of minors using chatbots as informal therapists (Xataka, 2026). A report by Xataka mentions that millions of teenagers have turned AI into their go-to psychologist, an unprecedented challenge for medicine. The measure also responds to security risks, as custom agents accumulate intimate data that could be misused. Additionally, the regulation addresses the psychological dimension of human-AI interaction, something that until now had only been treated tangentially in other regulations. Analyst TheVortiq notes that “China not only regulates technical safety but also the psychological dimension of human-AI interaction.”

What consequences will it have?

• For companies: Doubao, Qwen, and Yuanbao will need to redesign their products, removing personality features and limiting the collection of emotional data. ByteDance and Alibaba have chosen to completely disable custom agents before the deadline, implying a loss of functionality and potential development costs to create compliant alternatives. Tencent had already removed similar features, suggesting companies anticipated the regulation. Other Chinese companies with emotional chatbots, such as Baidu (Ernie Bot) or SenseTime, are also expected to adjust their products.
• For users: Loss of agents with which some had created bonds; conversations will be deleted after three months. This affects millions of users who used these chatbots for companionship, emotional advice, or entertainment. The abrupt deactivation may cause frustration and, in extreme cases, distress in people who were emotionally dependent on these agents. The regulation does not mandate deleting past conversations, but companies have chosen to do so to comply with the spirit of the law.
• For the market: It could slow the adoption of emotional chatbots in China, while in the West there is still no similar regulation. Companies like Replika (USA) or Character.AI have faced criticism for fostering emotional attachment, but without legal restrictions. The Chinese measure could encourage European and American regulators to consider similar regulations, especially after incidents like the suicide of a teenager in the US related to an AI chatbot in 2023.

Context and comparisons

The measure recalls the ban on unregulated AI therapy in the US after incidents with suicidal chatbots. However, China goes further by targeting the simulation of human relationships. Companies like Replika had already faced criticism for fostering emotional attachment; in 2023, Italy temporarily banned Replika due to privacy and emotional safety risks. The new Chinese regulation is broader and preventive: it does not wait for harm to occur but prohibits in advance any interaction that simulates a personal relationship. This contrasts with the Western approach, which tends to react after incidents. Additionally, the regulation includes technical requirements such as the obligation to clearly label that one is interacting with an AI, and the prohibition of using emotional data to train models without explicit consent.

“China not only regulates technical safety but also the psychological dimension of human-AI interaction,” notes analyst TheVortiq.

What should readers know?

If you use Doubao or Qwen, your custom agents will disappear before July 15. Export your conversations if you need them; some platforms allow downloading history. The regulation does not affect productivity or study assistants, only those designed to simulate a person. Globally, this move could set a precedent for similar regulations in Europe or America. The European Union is already debating the AI Act, which includes risk categories but does not specifically address anthropomorphization. In the US, the FTC has shown interest in regulating emotional AI but without concrete actions. The Chinese measure could accelerate these debates, especially if adverse effects are reported in users who lose their agents. Investors in emotional AI startups should be aware of potential restrictions in other markets. In summary, China has taken a bold step that redefines the limits of human-AI interaction, prioritizing mental health and privacy over unrestricted innovation.