The Guidelines for Deployment and Application of Large AI Models in Government Affairs, jointly issued by the Cyberspace Administration of China and the National Development and Reform Commission, marks China's first specialized policy for large model applications in government sectors. It outlines measures concerning application scenarios, standardized deployment, and operational management, signaling the entry of AI large model applications in government affairs into a new phase of orderly advancement.

I. Current Status of AI Large Model Applications in Government Services

In recent years, AI large model technology has evolved from classification to generation, from single-modal to multi-modal, and from specialized to general-purpose applications, becoming a vital engine for high-quality economic and social development. Estimates indicate that China's AI industry scale will exceed 900 billion yuan in 2024, with over 5,300 related enterprises by September 2025, accounting for 15% of the global total and forming a complete industrial ecosystem.

The government sector represents a key application front for large models. Following breakthroughs in domestic open-source large model technology in 2025, various local governments and departments proactively adopted these technologies, accelerating implementation. Large models are already deployed in internal operations and public services, with “text + chatbot” solutions diverting nearly one-third of manual services, while intelligent task dispatch and pre-review processes replacing nearly half of manual work. Applications are expanding into areas like intelligent drafting assistance, driving greater efficiency in government services. The State Council's earlier “Opinions on Deepening the Implementation of the ‘AI+’ Initiative” also emphasized AI adoption in governance, and the Guidelines further standardize and guide innovative practices.

II. Four Key Innovation Directions of the Guidelines

(1) Empowering Through Technological Innovation

The Guidelines leverage technological innovation to advance modern government governance. They highlight the advantages of large language models in complex semantic understanding and multimodal generation, integrate technologies like knowledge graphs and reinforcement learning with government service scenarios, and encourage exploration of emerging applications such as intelligent government entities and embodied intelligence.

(2) Scenario-Driven Innovation

The Guidelines focus on common, high-frequency government needs, proposing 13 specific scenarios across four major categories including public services and social governance. They clarify the technical capabilities, processes, and outcomes for each scenario. The Guidelines also permit scenario exploration tailored to practical contexts, requiring the summarization and promotion of exemplary cases to foster an innovation ecosystem.

(3) Operational Model Innovation

Addressing the complexity of deploying and maintaining large models, the Guidelines emphasize full-chain collaboration—such as leveraging media review for training data—and establishing a dynamic, end-to-end update mechanism supported by technology, data, and user feedback to ensure sustainable application implementation.

(IV) Governance Mechanism Innovation

The Guidelines clarify the “supporting” role of large models, mandating strict content review for public-facing applications to mitigate “hallucination” risks. They advance the classification and tiered management of government large model data, rigorously upholding security standards.

III. Driving High-Quality Development of “AI + Government Services” Through the Guidelines

(1) Upholding Coordinated Deployment

Advance unified planning and reuse of large model capability foundations based on the principle of “systematic planning and shared construction,” addressing the needs of eastern, central, and western regions as well as diverse groups to ensure smart and inclusive services.

(2) Strengthen Demonstration Leadership

Validate technologies through real-world scenarios, deepen collaboration among government, industry, academia, and research, eliminate formalism, and ensure large models genuinely serve the public.

(3) Fortify Security Boundaries

Establish closed-loop management and monitoring mechanisms throughout the entire process, implement requirements for lawful data and algorithmic documentation, and unleash innovative vitality within a secure framework.