On May 9, 2026, China witnessed a密集发布 of significant AI governance policies. The Ministry of Industry and Information Technology officially launched the AI science and technology ethics review pilot program, while the Cyberspace Administration, National Development and Reform Commission, and Ministry of Industry and Information Technology jointly issued the Implementation Guidelines for the Standardized Application and Innovative Development of AI Agents, marking China’s AI governance entering the substantive implementation phase from framework design.

1. Ministry of Industry and Information Technology Launches AI Ethics Review Pilot

1.1 Policy Background

The Ministry of Industry and Information Technology recently issued a notice officially launching the Artificial Intelligence Science and Technology Ethics Review and Service Pilot Program. This program relies on provinces where national AI industry innovation application pilot zones are located, taking the lead in exploring the implementation path of AI science and technology ethics review and services.

Previously, the Ministry and ten departments jointly issued the “Artificial Intelligence Science and Technology Ethics Review and Service Measures (Trial)”, and this pilot program aims to promote the implementation of these measures in Beijing, Shanghai, Guangzhou, Shenzhen, Hangzhou, Chengdu and other first-batch cities.

1.2 Key Tasks

The pilot program has clarified the following key tasks:

Task Area Specific Content
System Building Refine provincial AI science and technology ethics review system standards, guide the construction of science and technology ethics committees
Service System Explore building ethics review and service centers
Practice Exploration Carry out AI science and technology ethics review practices
Standard Conversion Transform review practice experience into technical standards
Monitoring and Early Warning Establish a national AI science and technology ethics risk monitoring service network

1.3 Involved Fields

Cities focus on the following vertical application fields for ethics review:

  • Foundation: Large models, computing platforms, training data
  • Manufacturing: Industrial AI, robotics
  • Education: AI-assisted teaching, smart campuses
  • Healthcare: AI diagnosis, drug development
  • Finance: Intelligent risk control, quantitative trading
  • Agriculture: Smart agriculture, AI breeding

1.4 Risk Scenarios

Currently focused on investigating ethics risks include:

  • Algorithm Discrimination: Gender, age, and regional discrimination in AI decision-making
  • Emotional Dependence: Impact of AI companion applications on minors and the elderly
  • Privacy Leakage: Personal information protection in training data
  • Responsibility Attribution: Determination of liability when AI causes damage

2. Three Departments Jointly Issue AI Agent Specification Implementation Guidelines

2.1 Policy Overview

The State Internet Information Office, National Development and Reform Commission, and Ministry of Industry and Information Technology jointly issued the “Implementation Guidelines for the Standardized Application and Innovative Development of AI Agents”. This is the first systematic regulatory document for AI Agents (Agent) in China.

2.2 Basic Principles

The “Implementation Guidelines” clarify four basic principles for AI Agent development:

  1. Safety and Controllability: Ensure predictable AI Agent behavior and manageable risks
  2. Standardization and Order: Establish unified access, operation, and exit mechanisms
  3. Innovation Driven: Encourage original technology breakthroughs
  4. Application Led: Drive technology iteration through actual application scenarios

2.3 Four Major Initiatives

The “Implementation Guidelines” propose specific initiatives from four dimensions:

(1) Consolidate Development Foundation

  • Improve technical foundation
  • Build standards and protocols
  • Promote interoperability

(2) Ensure Safety Bottom Line

  • Clarify product standards
  • Prevent security risks
  • Improve governance system
  • Strengthen industry self-regulation

(3) Strengthen Application Leadership

Promote around 19 typical application scenarios:

Field Application Scenario
Scientific Research AI for Science, Materials Discovery, Drug Design
Industry Development Smart Manufacturing, Supply Chain Optimization, Quality Testing
Consumption Boost Smart Recommendations, Virtual Try-on, Personalized Marketing
People’s Livelihood Smart Healthcare, Online Education, Elderly Care
Social Governance Smart Cities, Traffic Scheduling, Emergency Response

(4) Build Innovation Ecosystem

  • Promote industrial cooperation
  • Strengthen application promotion
  • Cultivate benchmark enterprises

3. AI Terminal Intelligence Classification National Standard

3.1 Standard Architecture

The Ministry of Industry and Information Technology, Ministry of Commerce, and State Administration for Market Regulation jointly launched the “Artificial Intelligence Terminal Intelligence Classification” series national standards.

This standard adopts a “2+N” architecture:

  • “2”: Basic standards
    • “Part 1: Reference Framework”
    • “Part 2: General Requirements”
  • “N”: Vertical category standards (mobile phones, computers, televisions, automobiles, etc.)

3.2 L1-L4 Classification System

Level Name Capability Description
L1 Responsive Passively respond to user instructions
L2 Tool-based Complete tasks by calling tools based on context
L3 Assisted Proactively provide suggestions, assist decision-making
L4 Collaborative Deep collaboration with users, autonomous planning (to be perfected later)

4. Industry Impact Analysis

4.1 Impact on AI Enterprises

Impact Dimension Specific Manifestation
Compliance Cost Enterprises need to establish ethics review mechanisms
Product Design Need to consider L1-L4 classification requirements
Data Governance Higher requirements for training data quality
Overseas Compliance Domestic standards may become international reference

4.2 Impact on Developers

  • Technology Selection: Prioritize tools and frameworks that comply with standards
  • Code Standards: AI-generated code needs to meet safety guidelines
  • Continuous Learning: Follow policy developments and adjust direction in a timely manner

4.3 Investment Opportunities

Policies benefit the following segmented tracks:

  1. AI Security Companies: Content moderation, risk detection
  2. Compliance Services: Data governance, privacy computing
  3. Standardization Services: Evaluation, certification
  4. Industry Solutions: Vertical field compliance implementation

5. International Comparison

Country/Region Governance Characteristics
China Dual-track system (ethics review + agent specification), government-led
EU AI Act, risk-based classification
US Industry self-regulation, supplemented by executive orders
UK Flexible regulation, sandbox pilots

The speed and systematic nature of China’s policy release is relatively rare internationally, demonstrating the high importance regulatory authorities attach to AI governance.

6. Conclusion and Outlook

May 9, 2026 can be called China’s AI governance “Policy Day”:

  • Ethics Review Pilot Program solves the problem of “whether it exists”
  • Agent Specification solves the problem of “how to do it”
  • Terminal Classification Standards solves the problem of “how to measure”

It can be foreseen that with the implementation of these policies:

  1. The AI industry will accelerate into standardized development
  2. AI products that do not meet ethics standards will face rectification or removal
  3. Compliance capability will become the core competitiveness of AI enterprises
  4. A number of “AI + Compliance” entrepreneurial opportunities will emerge in vertical fields

As AI practitioners, we need to pay close attention to the introduction of policy details and layout compliance capabilities in product design and business development in advance.

Further Reading:

  • “Artificial Intelligence Science and Technology Ethics Review and Service Measures (Trial)”
  • “Implementation Guidelines for the Standardized Application and Innovative Development of AI Agents”
  • “Artificial Intelligence Terminal Intelligence Classification” series national standards (draft for comments)

Related Reading: