China releases AI Security Governance Framework 3.0

The National Information Security Standardization Technical Committee has released the ‘Artificial Intelligence Security Governance Framework 3.0’. This updated version marks a significant shift by establishing a dedicated chapter for AI agents (intelligent agents), addressing the unique security risks they pose as they move from laboratory settings to large-scale industrial applications in sectors such as finance, manufacturing, and energy.

The new framework introduces standardized requirements for agent security, including identity and permission management, multi-layer workflow interception with mandatory human approval for high-risk actions, supply chain security for external plugins, and full-link monitoring and auditing. These measures aim to mitigate risks such as unauthorized operations, data leakage, and malicious plugin injections.

Parallel to these regulatory updates, concerns have been raised within the scientific community. A group of 25 Fields Medalists has issued a warning regarding the impact of AI on mathematical research, cautioning that treating complex problem-solving as a mere performance metric for AI models may undermine the integrity of scientific inquiry and the depth of human understanding.