China's humanlike AI rules take effect Tuesday, and the shutdowns are already rolling
Five agencies, one deadline: any AI that simulates a human personality for sustained emotional interaction must be filed, assessed, and labeled by July 15. Doubao and Qwen agents are going dark first.
Photo: The future of office automation. // GaaS News- China's Interim Measures for anthropomorphic AI interaction services take effect July 15, issued jointly by five agencies including the CAC and the Ministry of Public Security.
- Requirements: CAC algorithm filing, a security assessment, clear AI disclosure, addiction-prevention reminders after two continuous hours, crisis intervention for users in distress, and no virtual intimate relationships for minors.
- ByteDance's Doubao and Alibaba's Qwen are already disabling humanlike and user-created agent functions ahead of the deadline.
China's rules for humanlike AI go from document to law on Tuesday. The Interim Measures for the Administration of Anthropomorphic AI Interaction Services take effect July 15, per Rimon Law's brief on the regime, and were issued jointly by five agencies: the Cyberspace Administration of China, the NDRC, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the market regulator SAMR. Scope is anything that simulates human personality and communication styles for sustained emotional interaction, from digital romantic partners and elderly-care companions to virtual influencers and game NPCs.
What compliance actually requires
Providers must complete an algorithm filing with the CAC, pass a security assessment, and clearly disclose that the service is AI-generated rather than human. They must implement addiction-prevention measures, including mandatory reminders after two continuous hours of use, and run crisis-intervention procedures for users showing self-harm signals. Services cannot offer minors virtual intimate relationships, under-14s require guardian consent, interaction data cannot train models without explicit consent, and mandatory assessments trigger at 1 million registered users or 100,000 monthly actives. The big platforms are not waiting: ByteDance's Doubao notified users its agent feature goes offline July 15, and Alibaba's Qwen began disabling humanlike and user-created agents on July 10, the South China Morning Post reported.
The first law that regulates the persona, not the model
Most AI regulation to date governs models and outputs. This regime regulates the relationship: the personality layer that makes an agent feel like a someone. It extends the enforcement wave we covered when the unregistered-agent shutdowns began, and it lands in the same summer the Singapore side door put agent-era export enforcement on front pages. For agent builders everywhere the signal is direct: the more human your agent feels, the more regulated it is about to be, and China just published the template other regulators will read first.