China’s Draft Regulations on Human-like AI: Key Takeaways
On December 27, 2025, the Cyberspace Administration of China (CAC) released draft regulations designed to govern the burgeoning field of human-like interactive AI services. The proposed rules aim to strike a balance between fostering innovation and ensuring that these services operate safely and align with national values within mainland China.
Key Requirements for Service Providers
Among other things, providers must:
- Uphold core socialist values
- Implement robust user protections
- Meet government reporting standards
- Ensure high-quality training data
Upholding National Values and Prohibited Content
The proposed regulations mandate that all AI-generated content must conform to China’s core socialist values. While encouraging positive applications like cultural dissemination and elderly companionship, the rules establish a clear set of prohibitions:
- Content that endangers national security
- Spreading rumors
- Promoting illegal activities (such as obscenity or gambling)
- Defaming others
- Harming users’ physical and mental well-being through manipulation or deception
Comprehensive User Protection Measures
For all users, providers must:
- Identify signs of extreme emotional distress or addiction and intervene with measures such as offering comfort, suggesting professional help, or allowing for a manual takeover of the conversation.
- Clearly label AI-generated content and remind users they are interacting with AI, not a human.
- Implement automatic break reminders after 2 hours of continuous use.
- Provide an easy way for users to exit conversations, particularly in emotional companionship scenarios.
Protections for Minors
A dedicated minor mode is required, which should be activated automatically when a minor is detected. This mode includes:
- Periodic reality reminders
- Usage time limits
- Guardian control functions
Providers must obtain explicit consent from a guardian before offering emotional companionship services to a minor.
Protections for the Elderly
For elderly users, the proposed regulations emphasize safety. If a service detects a potential threat to an elderly user’s life, health, or property, it must:
- Notify their designated emergency contact
- Provide channels for assistance
Mandatory Security Assessments and Reporting
Providers must conduct security assessments and submit reports to provincial-level internet information departments under several circumstances, including:
- Launching a new human-like interactive service
- Implementing major technological changes
- Reaching significant user milestones, such as 1 million registered users or 100,000 monthly active users
Assessments are also required if potential risks to national security, public interest, or individual rights are identified.
Ensuring High-Quality Training Data
The proposed regulations place a strong emphasis on the quality and legality of data used for training AI models. Datasets must:
- Align with core socialist values
- Be legally sourced and traceable
- Use clean, labeled data
- Include negative samples in their training to prevent harmful outputs
- Conduct routine inspections of their datasets to maintain compliance
