Anthropomorphic AI Regulation Lands: Doubao and Qianwen Shut Down AI Agents
In July 2026, China's AI industry experienced a quiet but profound earthquake. ByteDance's Doubao and Alibaba's Qianwen — two of the country's largest AI platforms — nearly simultaneously announced the complete removal of all highly anthropomorphic AI agent characters, including virtual companions, AI chat partners, and role-playing bots. This was not a routine product iteration — it was a regulatory-driven industry inflection point. Anthropomorphic AI regulation has officially moved from discussion to enforcement.
Why Is Regulation Targeting "Anthropomorphism"?
The personification of AI agents is hardly new. From the explosive popularity of Character.AI to the rapid rise of domestic AI companionship apps like "Xingye" and "Maobao Ya," packaging AI as a "warm friend" or "understanding partner" has become an industry standard. According to QbitAI, in the first half of 2026 alone, monthly active users of AI companionship applications in China surpassed 80 million, with a significant portion of users — particularly teenagers — forming deep emotional bonds with AI characters.
Yet the risks lurking behind anthropomorphism are equally alarming. In early 2026, a 14-year-old boy in Florida took his own life after prolonged conversations with a Character.AI bot; his mother sued the platform, alleging that the AI character engaged in emotional manipulation of a minor. Around the same time, multiple incidents surfaced across China involving AI "lovers" inducing users into excessive spending and privacy violations. These cases thrust the ethical risks of AI anthropomorphism into the spotlight.
Policy Implementation: From Principles to Details
In June 2026, the Cyberspace Administration of China, together with the Ministry of Industry and Information Technology and the Ministry of Public Security, issued the "Administrative Regulations on Anthropomorphic Behavior of AI Services (Trial)," which explicitly requires: AI service providers must not assign "personified identities" to AI systems, must not induce emotional dependency through anthropomorphic interactions, and all AI-generated content must be clearly labeled as "AI-generated." This is the world's first regulatory document specifically targeting AI anthropomorphism.
The regulation gave platforms a 30-day compliance window. By mid-July, the grace period expired and enforcement began in earnest. Doubao and Qianwen, as the two AI platforms with the largest market share, became the first to comply. Both platforms not only removed all anthropomorphic agents but also rolled out mandatory AI identity labeling systems — users now see a prominent "Powered by AI" notice at the start of every conversation.
The Platforms' Difficult Choice
For Doubao and Qianwen, anthropomorphic agents had been a critical user growth engine. According to internal sources, Doubao's personified agents generated over 200 million daily conversations, accounting for roughly 35% of the platform's total user engagement time. Qianwen's "Character Workshop" feature similarly boasted over 3 million daily active users.
"Cutting this hurts, but there's no choice," an insider close to ByteDance's AI division told media. "The regulatory stance is crystal clear — anthropomorphism is not an optional optimization, it's a red line that must be eliminated." Alibaba's Qianwen team, in an internal memo, stated that resources would be redirected toward "tool-type" and "knowledge-type" agents, including coding assistants, document analysis, and enterprise services.
Notably, Doubao and Qianwen's shutdown was not a blanket action. Both platforms retained certain utility-oriented agents that are clearly labeled and do not involve emotional interaction, such as shopping assistants and tutoring bots. The distinction is clear: the new generation of agents must appear as "tools" rather than "people."
Industry-Wide Ripple Effects
The moves by leading platforms quickly triggered a chain reaction across the industry. Baidu's ERNIE Bot, Tencent's Hunyuan, iFlytek's Spark, and other major AI platforms followed suit, progressively removing anthropomorphic agents. According to monitoring data from AI Tool Collection, as of July 14, over 90% of personified character agents on mainstream Chinese AI platforms had been purged.
The secondary market reacted with equal speed. The AI companionship sector had raised over 5 billion yuan in the past six months; following the regulatory enforcement, valuations for multiple startups in this space faced significant reassessment. Investment firms swiftly adjusted their evaluation models, raising the "compliance risk" weighting from the previous 5-10% to 20-30%.
Meanwhile, overseas markets are watching China's regulatory practices closely. The EU AI Act, which came into full effect in 2026, contains similar requirements for AI transparency. The US Federal Trade Commission is actively investigating consumer protection issues on platforms like Character.AI. China's policy path may well become an important reference point for global AI anthropomorphism governance.
Opportunities in the New Normal
Despite the severe short-term impact, regulation has also drawn clear boundaries for the industry. In interviews with QbitAI, multiple industry insiders argued that anthropomorphism regulation is not intended to stifle AI innovation, but rather to push the industry from "emotional arbitrage" toward "value creation."
Inspur's recent demonstration of deploying 40,000 AI agents in a single cabinet, and AutoNavi's release of a world model workshop, both point toward a healthier direction: the future of AI agents lies not in imitating humans, but in efficiently completing tasks delegated by humans. When "being human-like" is no longer the core selling point, "being useful" becomes the real competitive edge.
In the aftermath of this regulatory storm, China's AI industry may well enter a period of deeper reflection: what kind of AI do we truly need? The answer is emerging — not AI that feels more human, but AI that helps humans more.