Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents
Issued by
Cyberspace Administration of China, National Development and Reform Commission, and Ministry of Industry and Information Technology
- September 30, 2026 · Correction — Corrected the issuing agencies, date (8 May 2026) and legal status: this is a policy framework, not a binding regulation. (Cody Maxwell)
- September 30, 2026 · Correction — Rewrote the entry to match the published Opinions: a non-binding policy framework with no effective date, no penalties and no enumerated sensitive sectors (fact-check finding). (Cody Maxwell)
- September 30, 2026 · Correction — Checked the rewrite against the CAC release and removed one unsourced comparison from the FAQ. (Cody Maxwell)
- October 1, 2026 · Correction — Rewrote the practical steps, newsletter hook, search description, and audience fields to match the entry's corrected content. (Cody Maxwell)
Three Chinese regulators released these Implementation Opinions on 8 May 2026, China's first policy framework dedicated to AI agents. The document is a policy framework rather than a binding regulation. It sets principles on user control and agent authority, with filing, compliance testing, and recall expected in sensitive sectors.
Applies To
Overview
The Cyberspace Administration of China, the National Development and Reform Commission, and the Ministry of Industry and Information Technology released these Opinions on 8 May 2026. They implement the State Council's AI+ initiative and treat AI agents as a distinct category: systems capable of autonomous perception, memory, decision-making, interaction, and execution. The document is a policy framework, not a promulgated regulation, and it has no separate effective date. It calls for governance graded by application scenario and potential impact. For sensitive fields and key industries, cyberspace and sector regulators will decide which scenarios to open and apply measures such as filing, testing, and recall of problem products. Lower-risk uses rely on self-testing, information reporting, platform management, and industry self-discipline. It also sets product principles on decision authority: what only users may decide, what needs their authorization, and what agents may decide alone. Users keep the right to know and the final say over agents' autonomous decisions. The document names 19 typical application scenarios and calls for follow-on standards. It sets no fines or other sanctions.
Key Requirements
- •Grade agent governance by application scenario and potential impact.
- •In sensitive fields and key industries, expect regulators to set filing, testing, and recall measures for agents.
- •Define which decisions only the user may make, which need user authorization, and which an agent may make alone.
- •Keep users informed and give them the final say over agents' autonomous decisions.
- •For lower-risk uses, rely on self-testing, information reporting, and platform and industry self-governance.
- •The Opinions set no penalties of their own; binding duties come from existing laws and later measures.
What Your Organization Must Do
- →Map which of your AI agents operate in China's sensitive fields or key industries.
- →Watch for sector regulators deciding which agent scenarios they will open.
- →Document now which decisions only users may make and which need user authorization.
- →Build user notice and override controls so users keep the final say on autonomous decisions.
- →Prepare for possible filing, testing, or product recall measures in sensitive sectors.
- →Track follow-on standards and existing laws, including the generative AI measures, for binding duties.
Playbook Guidance
Step-by-step implementation guidance for compliance teams.
Frequently Asked Questions
- Are China's AI agent Implementation Opinions a binding regulation?
- No, they are an implementation opinion, a policy document that sets principles and tasks. They have no separate effective date and set no penalties, but they signal how regulators will govern agents.
- How do the Opinions define an AI agent?
- They define an agent as an intelligent system capable of autonomous perception, memory, decision-making, interaction, and execution.
- Do the Opinions require filing agents with the Cyberspace Administration of China?
- Not directly; for sensitive fields and key industries, cyberspace and sector regulators will decide which scenarios to open and apply measures such as filing, testing, and recall under existing laws.
- What do the Opinions say about decision authority?
- They call for clear boundaries between matters only the user may decide, matters needing user authorization, and matters an agent may decide alone. Users keep the right to know and the final decision.
- Do the Opinions apply to foreign companies?
- They set policy for agents used in China, so foreign companies offering agents there should follow how regulators implement them. Binding duties come from existing laws such as the generative AI measures.
- Is 15 July 2026 the effective date of the Opinions?
- No, that date belongs to China's separate interim measures on anthropomorphic AI interaction services. The agent Opinions were issued on 8 May 2026 with no deferred effective date.