On September 15, 2025, the AI Safety Governance Framework 2.0 (hereinafter referred to as “the Framework 2.0”) of China was issued by the SAC/TC260 (Cybersecurity) during the National Cybersecurity Week. It is considered a significant move to shape the future of artificial intelligence, signaling a major evolution in the country's strategy to manage the risks and opportunities presented by AI. 

The Framework 2.0 moves beyond its 2023 predecessor by introducing a more sophisticated system for categorizing AI-related dangers and establishing a tiered response protocol. The goal is to weave safety and oversight into the very fabric of AI systems, from their initial design to their final decommissioning. 

While the first framework identified core risks like algorithmic instability and data breaches, Version 2.0 casts a wider net. It formally acknowledges a new class of threat: "Derivative Risks." This category confronts the large-scale societal consequences of widespread AI adoption, such as:

·       Significant workforce displacement

·       Intensified pressure on energy grids and natural resources

·       The entrenchment of societal biases through automated systems

·       Potential negative impacts on educational integrity and innovation pipelines

A pivotal advancement in the new framework is the shift from simply classifying AI systems to grading the risks they pose. This risk-based approach assesses threats based on the specific context of use, the system's level of autonomy, and its potential scale of impact. Risks are now sorted into a five-tier hierarchy: i) low; ii) moderate; iii) considerable; iv) major; v) extremely serious. This nuanced grading system is designed to enable more precise and proportionate safety measures, ensuring that regulatory scrutiny matches the actual level of risk. 

Perhaps the most profound operational change is the adoption of a full life-cycle governance model. The previous version offered guidance to various stakeholders (e.g., developers, users). Framework 2.0 restructures this entirely, mapping detailed technical and compliance recommendations onto each phase of an AI system's existence—from data collection and model training to deployment, monitoring, and eventual retirement. This approach aims to eliminate regulatory gaps and align governance with the real-world journey of AI products. 

The introduction of these graded principles is expected to catalyze the creation of new national and industry-specific standards. Technical committees, including TC260, are already mobilizing expert groups to develop supporting documents, such as a standard on "Classification and Grading Methods for Artificial Intelligence Application."

Collectively, these updates illustrate China's ambition to construct a robust and self-reinforcing AI governance ecosystem. Framework 2.0 doesn't just revise the old rules; it builds a more detailed and dynamic regulatory architecture intended to keep pace with the rapid evolution of artificial intelligence.

BESTAO presents free monthly report on China compliance. It offers a comprehensive solution on observing various standards and regulatory activities in China. Sample of the monthly report please refer to:

https://www.bestao-consulting.com/detail?id=1740&status=bestao_library

Subscribe the free monthly report by register as a BESTAO website member at: https://www.bestao-consulting.com/login, or write an email to assistant@bestao-consulting.com.