Skip to content
NewsRegulation

China's first AI persona law takes effect July 15, forcing ByteDance and Alibaba to kill humanlike agents

· by Pondero Newsdesk

The short version

The Cyberspace Administration of China's Interim Measures on AI anthropomorphic interaction services take effect July 15. ByteDance's Doubao and Alibaba's Qwen are disabling all user-created AI agents. Data deletion follows by October 15.

China's first AI persona law takes effect July 15, forcing ByteDance and Alibaba to kill humanlike agents

Nine days from now, every user-created AI agent on ByteDance's Doubao and Alibaba's Qwen goes dark. The deadline is not a technical glitch or a business pivot. It is compliance with the world's first binding regulatory framework for AI services that simulate human personality, and neither company found a way to retrofit its existing agent architecture to meet the rules in time.

What the law requires

The Interim Measures for the Administration of AI Anthropomorphic Interaction Services were co-issued in April 2026 by the Cyberspace Administration of China and four partner agencies: the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation. They take effect July 15, per TechNode.

The rules target a specific category: AI services that simulate human personality traits, thinking patterns, and communication styles to provide sustained emotional interaction. Standard productivity chatbots, customer service systems, knowledge Q&A, and educational tools are explicitly excluded, provided they do not create ongoing emotional bonds with users.

For services that do fall inside the definition, the compliance bar is steep. Platforms must clearly disclose to users that they are interacting with AI rather than a human. They must implement dynamic anti-addiction prompts after two continuous hours of interaction, install instant-exit mechanisms that the platform must honor immediately, and detect signs of over-dependence in real time. Providers must also run security assessments and register their algorithms with provincial-level cyberspace authorities once registered users reach one million or monthly active users reach 100,000. Virtual intimate relationships with users under 14 are prohibited without explicit parental consent.

The cited risks in the measures span extremist content, privacy leaks, harm to mental health, and addiction.

Why the architecture failed

Both Doubao and Qwen built agents that are fundamentally incompatible with these requirements, per TechTimes. Each agent in Doubao carries a user-configurable identity, communication style, area of expertise, and a private memory store that persists across sessions. A user could maintain a research assistant, a creative writing partner, and a personal coach side by side, each with distinct memory that builds over time. That is precisely the sustained emotional scaffolding the regulation targets.

The compliance problem is structural. An agent whose core value is long-term continuity and relationship-building cannot simultaneously install the mandatory friction designed to interrupt that same bond. Rebuilding the feature from scratch under a new compliance architecture was judged more practical than retrofitting what already existed. Both companies reached the same conclusion independently.

What happens to your data

The two platforms chose different shutdown schedules and different data policies, per South China Morning Post.

ByteDance told Doubao users their agent features would go offline on July 15. After that date, users retain read-only access to agent configurations and chat histories until October 15, 2026. After October 15, ByteDance says the data will be processed in accordance with its privacy policy and will no longer be accessible or recoverable inside the app. The company is directing users to export content via screenshots or text sharing before the shutdown.

Alibaba moved faster. Qwen announced that humanlike interactive agents and user-created agents would be disabled on July 10, five days before the legal deadline, with broader agent functions offline by July 15. Alibaba has confirmed that agent configurations and conversation histories will be permanently deleted following the shutdown. No migration path was announced.

Tencent already pulled a similar feature from its Yuanbao assistant in June, per The Next Web, making the July 15 deadline a formality for that platform rather than a forcing event.

The regulatory contrast with Western markets

No equivalent regulation exists in the United States, the European Union, or the United Kingdom. OpenAI's persistent memory features, Anthropic's character-based Claude, and Grok's conversational voice mode face no requirement to disclose AI identity, limit interaction duration, or install anti-addiction systems. The EU AI Act addresses high-risk systems but does not specifically regulate AI companion or persona services the way Beijing's measures do.

Pan Helin, a member of an expert committee at China's Ministry of Industry and Information Technology, told the South China Morning Post that the policy prioritises safety, practical use, and standardisation. "Current agents are not yet mature," he said, per the SCMP report.

Beijing is not against agents as a category. The CAC issued guidance in May on the managed development of AI agents, and China released national standards in June covering agent identity, discovery, interaction, and tool use. The pattern is one of separating productive, task-focused agents from emotional companion AI, rather than blocking the technology outright.

For Western AI companies that operate in China, the regulation creates a dual-feature problem: maintaining a compliant, persona-stripped version for the domestic Chinese market while continuing to develop richer, persistent-personality features everywhere else. That compliance cost now falls equally on Chinese-origin companies operating domestically, which may reduce the competitive asymmetry but does not eliminate it.

What to watch next

Three things are worth tracking after July 15.

First, whether the Cyberspace Administration of China enforces the rules against platforms that have not proactively complied. The CAC has not signaled which companies it has reviewed beyond the major consumer apps.

Second, whether any Western AI company proactively adjusts its China-facing product offering rather than waiting for enforcement action. That decision would require disclosing that a jurisdiction-specific product split exists.

Third, whether the Interim Measures become a template for other governments. The EU has been examining emotional AI and AI companionship in the context of child safety. South Korea and Japan have active AI regulatory discussions. Beijing's rulebook is now the reference text for any regulator who wants to move in this direction.

The July 15 date is the legal effective date. It is also, for millions of Doubao and Qwen users, the last day their AI companions remember them.

Sources