Beijing Releases Comprehensive Policy Measures to Accelerate AI Agent Development

Beijing has officially released a comprehensive set of policy measures to accelerate the development of AI agents, jointly published by four municipal departments including the Beijing Development and Reform Commission, the Cyberspace Administration, the Zhongguancun Science Park Management Committee, and the Bureau of Economy and Information Technology.

The policy defines AI agents as intelligent systems with autonomous perception, memory, decision-making, interaction, and execution capabilities, positioning them as a critical new competitive frontier in artificial intelligence. The document notes that products such as “Xiaolongxia” and “Hermes” have rapidly gained traction, making agent technology a transformative force in production paradigms and lifestyles.

On the technology front, the measures adopt a comprehensive multi-track approach, simultaneously advancing foundational model capabilities and common agent framework technologies. Beijing plans to organize several key technology research projects leveraging local innovation advantages.

For applications, the policy emphasizes both native application development and scenario-driven deployment. Priority sectors include software and scientific research, where Beijing has strong foundations and growth potential. The government will create benchmark application scenarios to facilitate technology validation, product iteration, and real-world deployment.

A notable forward-looking element is the exploration of new economic paradigms: next-generation intelligent terminal products, OPC (One-Person Company) models, and Token-based economies. These are designed to restructure industrial value chains and unlock new growth drivers.

On safety governance, the policy mandates tiered and categorized regulatory oversight of AI agents, industry self-discipline, accelerated AI legislation, and the construction of safety infrastructure. The concept of “using models to govern models” is introduced, involving safety testing ranges and specialized security LLMs.

The measures also address key factor guarantees across computing power, financial support, data supply, and talent cultivation. Innovative financial instruments such as Token vouchers and agent service vouchers are proposed, alongside specialized training programs for AI agent talent.

Finally, the policy supports open-source development, including the construction of a China-SCO countries AI application cooperation center and the acceleration of globally influential open-source communities.

These comprehensive measures signal Beijing's systematic approach to capturing the AI agent opportunity, potentially serving as a blueprint for other Chinese cities and regions.