Artificial intelligence grading scale China: Navigating TC260’s new security risk framework
The Chinese document “Cybersecurity Technology – Methods for Classifying and Grading the Security of AI Applications” specifies how AI applications are to be categorised by risk type and subsequently graded according to the severity of the respective risk. Experts from Ecovis provide an explanation of the draft, which was published on 15 July 2026. The consultation phase is currently ongoing.
Who is the standard for
The draft speaks directly to organisations that build AI applications and those that run them. Both groups would use its methods whenever they carry out classification and grading work on their own systems.
Contact Person
Step 1: Classification of the risk types across three dimensions
To work out what kind of risk an application poses, the draft examines it from three angles:
- The setting in which the system operates – meaning the sector it serves. For example, finance, healthcare, education or public services.
- What is the application’s job? E.g., is it producing content, supporting decisions or controlling processes autonomously?
- How capable and self-directed is the AI system itself?
We can explain the draft on the classification and risk assessment of AI applications to you in detail and provide recommendations for action.
Richard Hoffmann, Lawyer, ECOVIS Rechtsanwaltskanzlei Richard Hoffmann, Ladenburg, Germany
Step 2: Grading by likelihood and impact
Once the risk types are identified, the draft calls for a combined judgment of two factors: how likely a risk is to materialise and how severe the consequences would be. The result falls on a five-level scale running from low, through general and relatively serious, up to major and extraordinarily major. Where several criteria apply to one application, the highest and strictest of them determines the overall level.
What companies should do now
For businesses developing or deploying AI in China, the draft is worth reading closely. A sensible first move is to compare existing systems against the proposed categories and levels and to consider how particular design and deployment decisions could push an application into a higher tier. Building risk assessment and documentation into daily AI operations now, and contributing to the consultation before it closes, should reduce compliance costs later, once the standard and any binding obligations that rest on it take effect.
For further information please contact:
