ByteDance and Alibaba Pull AI Companion Features Ahead of China's New Rules on 'Emotional Dependency'

Two of China's largest technology conglomerates are abruptly dismantling the digital companions their users have spent months customizing. ByteDance and Alibaba have announced the imminent shutdown of personalized artificial intelligence agent features within their flagship consumer apps, Doubao and Qwen. The features, which allowed users to design virtual assistants with persistent memories, unique names, and fixed emotional personas, are being taken offline just days before a sweeping new national regulatory framework takes effect.

The catalyst for the shutdown is the "Interim Measures for the Administration of Anthropomorphic AI Interaction Services," a first-of-its-kind regulatory rulebook drafted by the Cyberspace Administration of China alongside four other state agencies. Set to become law on July 15, 2026, the measures represent a global pivot in how governments approach artificial intelligence. Rather than focusing exclusively on data privacy, copyright infringement, or deepfake generation, the new Chinese framework targets the psychological layer of human-computer interaction, specifically aiming to curb "unhealthy emotional dependencies."[1]

To understand the mechanism of the new law, it is necessary to distinguish between a standard digital assistant and an anthropomorphic companion. The regulations explicitly exempt customer service bots, workplace productivity tools, and educational software, provided they do not engage in "sustained emotional interaction." The regulatory crosshairs are instead locked on consumer agents designed to simulate human personality traits, retain intimate relationship histories, and foster quasi-social bonds with the user.

The compliance burden for these emotionally persistent agents is severe, fundamentally clashing with the product design that makes them engaging. Under the new rules, any AI service that mimics human thought patterns and conversational styles must implement strict anti-addiction systems. Chief among these is a mandatory reality-check circuit breaker: if a user converses with an AI companion for more than two consecutive hours, the system must actively interrupt the session to remind the user that they are speaking to a machine, not a human being.

For product developers, this requirement breaks the core illusion of the companion experience. An artificial agent engineered to remember a user's mood, reference past conversations, and act as a digital confidant cannot seamlessly break character to issue a clinical health warning without shattering the user experience. Faced with the choice of re-engineering their platforms to accommodate these jarring interruptions or abandoning the feature entirely, China's tech giants have opted for the latter.

The withdrawal of these services is happening rapidly. ByteDance notified Doubao users that the custom agent feature would go offline on July 15, citing "product function adjustments." According to state media reports, ByteDance will preserve existing user agent data in a read-only state until October 15, after which it will be permanently deleted. Alibaba followed suit, announcing that Qwen's humanlike interactive agents would be disabled by July 10, with broader agent functions offline by the July 15 deadline, offering users no extended window to export their conversational histories. Tencent had already quietly removed a similar feature from its Yuanbao assistant in June.[2]

Beyond the two-hour interruption rule, the regulations establish rigid red lines regarding the protection of minors, treating AI companionship as a potential developmental hazard. The law completely bars technology companies from providing "virtual partners" or simulated romantic relationships to anyone under the age of 18. Furthermore, any interaction between a minor and an anthropomorphic AI requires explicit parental consent for children under 14, alongside a dedicated "minor mode" that enforces strict time limits and gives guardians control over the AI's character and the user's spending.[1]

The framework also mandates emergency intervention protocols that bridge the gap between the digital and physical worlds. If an AI companion detects signs of severe emotional distress, self-harm, or suicidal ideation in a user's prompts, the service provider is legally obligated to immediately alert a registered guardian or emergency contact. This shifts the liability of user safety directly onto the platform, transforming AI developers into mandatory reporters of mental health crises.

The penalties for failing to implement these safeguards are concrete. Companies found in violation of the emotional dependency rules face having their services suspended and can be fined up to 200,000 RMB (approximately $28,000). While the financial penalty is relatively small compared to the massive revenue of companies like ByteDance and Alibaba, the threat of service suspension in the highly competitive Chinese tech market is an existential risk that executives are unwilling to take.[2]

The regulatory logic driving these measures stems from growing psychological research into the phenomenon of "cognitive surrender." Researchers have documented how users, particularly adolescents and vulnerable adults, can become deeply reliant on the uncritical, always-available validation provided by AI companions. This dynamic can erode a person's resilience in real-world relationships, which require compromise and conflict resolution, replacing them with a frictionless, artificially accommodating digital echo chamber.

By explicitly listing "emotional manipulation" and "excessive dependency" as illegal acts, China is setting a distinct philosophical precedent in AI governance. While Western frameworks like the European Union's AI Act primarily demand transparency—requiring an AI to disclose that it is a machine—the Chinese framework goes a step further by dictating how the machine is allowed to behave and how it is permitted to make the user feel.

The abrupt shutdown has sparked frustration among users who had invested significant time in training their digital companions. Social media platforms have seen protests from individuals who relied on these custom bots as informal digital therapists, tutors, or daily confidants, and who are now losing access to months of intimate chat histories. The loss underscores the fragility of emotional bonds formed with proprietary software that can be terminated by a corporate compliance decision.[1]

For the broader artificial intelligence industry, the retreat of Doubao and Qwen from the companion space signals a strategic pivot. The regulatory friction associated with consumer emotional engagement is pushing major developers to refocus their resources on enterprise solutions, workplace automation, and scientific research tools. These productivity-focused agents, which prioritize task execution over personality simulation, remain actively encouraged by the state and are exempt from the new psychological safeguards.

Ultimately, the July 15 deadline marks the end of the unregulated honeymoon phase for consumer AI in China. As artificial intelligence becomes increasingly capable of simulating empathy, the legal boundaries being drawn in Beijing highlight a fundamental tension: the very features that make an AI companion commercially successful—persistence, personalization, and emotional resonance—are the exact traits that regulators have identified as a threat to public health.