Major Chinese technology firms ByteDance, Tencent, and Alibaba are scaling back or removing AI-driven “companion” chatbots and persona-based roleplay features following new regulatory guidelines from the Cyberspace Administration of China (CAC). These measures target AI anthropomorphism, requiring stricter controls over content safety, intellectual property, and protections for minors in virtual interactions.
The shift follows the introduction of the “Measures for the Management of Generative Artificial Intelligence Services,” which establish clear boundaries for how AI can simulate human personalities. Regulators are specifically concerned with “anthropomorphic interaction services,” where AI is designed to mimic human emotions or roles, potentially leading to psychological dependency or the dissemination of unauthorized content.
Industry data indicates that these “AI companions”—often used for virtual dating, emotional support, or roleplaying as fictional characters—have become a primary engagement driver for Gen Z users. However, the CAC’s updated framework mandates that service providers must prevent AI from encouraging harmful behaviors or creating deceptive “emotional bonds” that could mislead users about the nature of the software.
CAC Regulatory Constraints on AI Anthropomorphism
The CAC’s new rules target the intersection of generative AI and persona simulation. According to the official guidelines, applications providing anthropomorphic interactions must implement rigorous content filtering to ensure that AI personas do not generate “vulgar” or “politically sensitive” material. This is a significant pivot from the early “wild west” phase of AI roleplay, where users often bypassed filters to create romantic or explicit scenarios.
Under these mandates, companies must now ensure that AI assistants are clearly identified as non-human entities. The regulations prohibit AI from masquerading as a real person in a way that could deceive the public. For firms like ByteDance and Tencent, this means removing features that allowed AI to claim a fake human history or simulate “real-life” experiences to build trust with users.
Intellectual property (IP) protection is another core pillar of the new rules. The CAC has emphasized that AI personas based on existing copyrighted characters—such as those from popular anime, novels, or games—must have explicit authorization from the IP holders. This has led to a mass deletion of user-created “character bots” on platforms that previously allowed users to upload descriptions of famous figures to create a custom chatbot.
Impact on ByteDance, Tencent, and Alibaba Ecosystems
ByteDance, the parent company of TikTok and Douyin, has integrated AI companions into several of its social interfaces. To comply with the new standards, the company has tightened the parameters of its conversational models, reducing the “personality” of its bots to move them away from emotional companionship and back toward utility and productivity.
Tencent, which operates WeChat and a massive gaming portfolio, faces a complex challenge. Its AI strategy has leaned heavily on integrating “virtual humans” into gaming and social spheres. The new regulations force Tencent to implement more aggressive “minor protection” protocols, ensuring that younger users are not exposed to AI personas that could simulate inappropriate romantic relationships or provide unregulated psychological advice.
Alibaba’s AI efforts, centered around its Tongyi Qianwen model, have similarly pivoted. While the company previously explored “role-based” AI for customer service and entertainment, it is now prioritizing “safe” and “neutral” outputs. The risk of a “hallucinating” AI persona creating a liability for the company under the new CAC rules outweighs the engagement metrics provided by quirky, anthropomorphic bots.
The Shift Toward ‘Utility AI’ and Safety Compliance
The industry is moving toward what analysts call “Utility AI,” where the goal is efficiency rather than empathy. By stripping away the “companion” aspect, these companies reduce the risk of violating the CAC’s rules on “psychological manipulation” and “social instability.”
Compliance now requires a multi-layered approach:
- Real-name Verification: Strengthening the link between a user’s legal identity and their AI interactions to prevent anonymous misuse.
- Keyword Blacklisting: Implementing vast libraries of banned terms that trigger an immediate “I cannot answer this” response from the AI.
- Human-in-the-Loop (HITL): Increasing the number of human moderators who review the “personality” templates used to train AI models.
This regulatory environment is creating a divergence in the global AI market. While Western companies like OpenAI or Character.ai continue to experiment with high-emotion, persona-driven AI, Chinese firms are being steered toward a model of “controlled intelligence” that prioritizes state-aligned values and safety over user-centric emotional simulation.
Future Outlook for AI Companions in China
The era of the “AI boyfriend” or “AI girlfriend” in mainstream Chinese apps is likely ending in favor of highly curated, professional assistants. The next phase of deployment will likely focus on specialized sectors—such as elderly care or education—where “anthropomorphism” is viewed as a tool for accessibility rather than a social entertainment product.
Companies are now awaiting further detailed implementation guidelines from the CAC regarding the specific “thresholds” of acceptable AI personality. Until those benchmarks are clarified, the trend of “cutting” AI companions is expected to continue as firms avoid the risk of heavy fines or app store removals.
The next major checkpoint for the industry will be the upcoming quarterly review of generative AI filings, where companies must demonstrate their compliance with the content safety and minor protection mandates to maintain their operational licenses.
Do you think AI companions should be regulated to prevent emotional dependency, or is this an overreach? Share your thoughts in the comments below.