China Issues "Digital ID Cards" for AI Agents: The World's First National Agent Policy
On May 8, China's Cyberspace Administration, National Development and Reform Commission, and Ministry of Industry and Information Technology jointly released the "Implementation Guidelines for Standardized Application and Innovative Development of Intelligent Agents" — the world's first government-level dedicated agent governance document. The era of AI Agent "identity cards" has officially begun.
A Document That Redefines an Industry
A joint policy from three major ministries is rare in the AI space — the gravity of this move signals that agent governance has reached the highest levels of China's policy agenda.
1. The Government Defines "Intelligent Agent" for the First Time
The document provides the first official national definition: An intelligent agent is an intelligent system with autonomous perception, memory, decision-making, interaction, and execution capabilities — a vital form of AI product and service.
This matters because it elevates Agent from "a technical concept" to "a formally recognized product and service category." All subsequent regulation, standards, and support policies will be anchored to this definition.
2. 19 Named Scenario Categories
The document explicitly identifies 19 typical application scenarios where agent deployment should be actively promoted, including healthcare, transportation, public safety, and media.
70% adoption rate among large and medium-sized enterprises is an explicit quantitative target — this is not "suggesting you try it," this is "you should start moving."
3. Three-Tier Decision Authority
This is the most substantive regulatory framework — a three-tier decision authority system for agents:
| Level | Authority Scope | Example |
|---|---|---|
| L1 | User decision only | Agent can only provide advice; no executable actions |
| L2 | User-authorized execution | Agent proposes a plan; user approves before auto-execution |
| L3 | Autonomous agent decision | Within preset boundaries, the Agent can independently complete a closed loop |
Complementing this is a classified governance mechanism: high-risk scenarios (healthcare, finance) require registration; low-risk scenarios rely on enterprise self-regulation.
AIP Protocol: The "Digital ID" for Every Agent
The most technically substantive aspect of the document is its explicit push for AIP (Agent Interconnection Protocol) national standards.
What Is AIP?
AIP is a multi-centric communication protocol designed for massive AI Agent interconnection, led by the China Electronics Standardization Institute. Seven series of national standards have entered the approval stage and will be officially released soon.
What Problem Does It Solve?
Current Agent interconnection efforts center around two protocol camps — MCP and A2A — both with blind spots:
| Protocol | Core Capability | Missing Capability |
|---|---|---|
| MCP (Anthropic) | Tool invocation, data access | Identity recognition, multi-agent collaboration |
| A2A (Google) | Agent-to-agent communication | Trusted access, behavioral audit |
| AIP (China National Standard) | Full-chain coverage | — |
AIP fills six critical gaps: 1. Trusted Access — Who are you? Who authorized you? 2. Identity Authentication — Is your "ID" valid? 3. Capability Discovery — What can you do? How do I invoke you? 4. Interconnection Collaboration — How do multiple Agents coordinate? 5. Settlement & Transaction — How is task completion settled? 6. Behavioral Audit — Every operation is traceable
The "Digital Identity Card" Mechanism
AIP uses an OID (Object Identifier) system to issue a unique "digital identity card" to every Agent — internally referred to as AIC (Agent Identity Code).
This means: every Agent running online in the future will have a unique identity code, just like a human. This code links to the Agent's capability scope, permission level, behavior history, and accountability.
From "Corporate Self-Regulation" to "National Mandate"
The most profound significance of AIP: it elevates agent governance from "companies regulating themselves" to "national standard enforcement."
Before this, AI Agent safety and compliance depended entirely on corporate self-discipline. With AIP, every Agent must: - Register its identity before connecting to any network - Operate within the permission classification framework - Maintain auditable logs for every operation - Register in advance for high-risk scenarios
For Nizwo's locally deployed AI computers, AIP is actually a tailwind — because local deployment inherently complies with "data stays local" requirements. When AIP mandates identity authentication and behavioral auditing, cloud-based Agents face significantly greater data compliance pressure.
The Debate: Will Regulation Stifle Innovation?
As with any major policy, AIP's rollout comes with controversy.
Proponents argue: - The more capable Agents become, the more baseline rules are needed. Running without "identity cards" — who's accountable if something goes wrong? - Explicitly naming 19 scenarios gives the industry defined commercialization pathways - Rules first mean large enterprises will actually feel safe adopting Agent technology
Critics worry: - Will the registration mechanism devolve into an "approval system"? Do small companies still have a chance? - Can the 7 national standards keep pace with Agent technology's iteration speed? - Will over-regulation cause China's Agent industry to fall behind globally?
A neutral observation: the pace of regulation matters enormously. If AIP can balance "ensuring safety" with "encouraging innovation," it could become infrastructure-scale advantage for China's Agent industry — like TCP/IP protocols established the foundation before traffic could flow in the internet era.
In one sentence: AI Agents in China have received both a "birth certificate" and an "identity card." For the industry, certainty is everything — knowing the rules is what gives enterprises the confidence to go all-in.
The AI Frontier column continuously tracks the latest AI industry developments. Follow us to never miss an industry pivot.