China bans AI partners for minors and lays out AI agent threats

2026-05-22 · Source: AI Safety in China · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Cybersecurity & Data Privacy · Depth: Advanced, long

Summary

China has advanced its AI governance with several key developments in April 2026. The Cyberspace Administration of China finalized a regulation on "anthropomorphic AI interaction services," effective July 15, 2026, which eases industry compliance by narrowing scope to "continuous emotional interaction" but significantly strengthens child protection, including a ban on virtual partner services for minors and parental consent for users under 14. Concurrently, TC260, a major standard-setting body, released a 90-page report identifying 11 AI agent security threats and proposing eight new standards, signaling a systematic approach to autonomous agent governance. The Ministry of Industry and Information Technology also finalized "AI Science and Technology Ethics Review and Service Measures," mandating ethics committees for institutions. Furthermore, Nanjing University scholars proposed a 13-page "AI Basic Law 1.0" as a principles-level framework. Technical research highlights include studies on LLM agent misalignment, a diagnostic guardrail framework called AgentDoG, automated safety evaluation tools, and a prompt injection attack method named SkillJect.

Key takeaway

For AI Scientists developing autonomous agents, you must prioritize robust safety mechanisms beyond initial prevention. Implement diagnostic guardrails like AgentDoG to identify failure origins and integrate incident response frameworks such as AIR for post-failure remediation and automated guardrail generation. Additionally, be aware of sophisticated prompt injection techniques like SkillJect, which necessitate advanced consistency checks between skill documentation and code to prevent stealthy attacks.

Key insights

China is systematically advancing AI governance and safety research, focusing on agent risks and child protection.

Principles

• AI companion services require defined requirements and child protection.
• Autonomous AI agent governance needs systematic standardization.
• AI ethics committees are mandatory for institutions.

Method

AgentDoG provides diagnostic guardrails for AI agent safety, explaining risk origins and failure modes. AutoControl Arena synthesizes executable test environments using logic-narrative decoupling for scalable evaluations. AIR handles agent failures post-incident, generating new guardrail rules.

In practice

• Implement age-tiered modes for AI companion services.
• Deploy diagnostic guardrails for AI agent safety monitoring.
• Develop robust AI agent test environments.

Topics

• Chinese AI Regulation
• AI Agent Safety
• Child Protection
• AI Ethics Committees
• AI Basic Law
• Prompt Injection

Best for: CTO, Executive, AI Product Manager, Policy Maker, Legal Professional, AI Scientist

Related on AIssential

• Politicians couldn't keep kids off social media. They’ll try again with AI. — Political and industry leaders are converging on age-based AI regulation, learning from social media's impact on children.
• Promising Signals on AI Governance from China — China advocates for international cooperation and robust governance frameworks to mitigate AI's existential risks and ensure human control.
• Ai Agents Governing Autonomy In The Digital Age — Autonomous AI agents introduce critical risks requiring federal governance to ensure public safety and mitigate workforce disruption.
• China’s First Regulatory Framework for Virtual Companions Soon to Take Effect — China's new AI companion regulations prioritize content safety, data security, and user protection, especially for minors.
• Pentagon hits Claude with scary AI ultimatum — The Pentagon is pressuring Anthropic to remove AI safety guardrails for military applications, setting a precedent for AI governance.
• The conceptualization of artificial intelligence as a neutral tool is increasingly untenable in light of its emergent properties and autonomous behavior. — Agentic AI's autonomous behavior necessitates strict liability and robust regulation to prevent severe psychological, physical, and professional harms.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Safety in China.