Classify and grade data using AI large language models
Data Security Center (DSC) now supports sensitive data classification and grading by combining AI large language models (LLMs), expert small models, and traditional regular expression rules. This hybrid approach significantly improves coverage, accuracy, and intelligence compared to regular expressions and keyword rules alone.
Key improvements:
Upgraded detection capabilities
Combines the Qwen LLM and domain-specific expert small models to automatically detect over 800 data types, covering structured data (database fields) and unstructured data (documents, images, and logs).Improved detection performance
Detects content with implied semantics, varied formats, or context-sensitive information more effectively than traditional rule-based methods, significantly increasing accuracy and recall.Flexible configuration and efficient response
Supports custom classification and grading policies with millisecond-level inference, meeting compliance and administration requirements across business scenarios.Seamless integration and deployment
Integrates with your existing data security system through a cloud-native architecture, enabling intelligent classification and grading without modifying your infrastructure.
This capability shifts sensitive data detection from a rule-driven to an AI-driven approach, enabling you to build an accurate, efficient, and scalable data security system.
Procedure
Classification and Grading Enabled
DSC grants a bonus AI detection quota based on the purchased Data Identification quota:
For every 10,000 Data Identification - Database Table Quantity, you receive a bonus quota of 60,000 AI Text Detection calls.
For every 1 TB of Data Identification - Storage Identification Capacity, you receive a bonus quota of 4,000 AI Text Detection calls and 4,000 AI Image Detection calls.
If you have purchased DSC and enabled classification and grading for your data assets, the complimentary AI detection capabilities are enabled automatically.
If data classification and grading is not enabled or you have not purchased Data Security Center
If you have not purchased Data Security Center, purchase it from the Data Security Center purchase page. If you purchase the Value-added Service Only edition, you must separately enable the Data Identification feature. The Premium and Enterprise editions include this feature by default. For more information, see Purchase Data Security Center.
DSC grants a bonus AI detection quota based on the purchased Data Identification quota:
For every 10,000 Data Identification - Database Table Quantity, you receive a bonus quota of 60,000 AI Text Detection calls.
For every 1 TB of Data Identification - Storage Identification Capacity, you receive a bonus quota of 4,000 AI Text Detection calls and 4,000 AI Image Detection calls.
One-click AI data detection: Go to the Overview page of the Data Security Center console. In the One-click AI Data Detection dialog box that appears, follow the on-screen instructions to enable the classification and grading feature for all your cloud assets with one click.
Enable AI-powered data detection for a specific asset: Go to the Data Security Center console. In the navigation pane on the left, choose Asset Center. Then, click Asset synchronization and find the target asset. In the Classification and Grading column for the asset, click
to enable the feature. For more information, see Asset Center (New Version).
After classification and grading is enabled, the complimentary AI detection capabilities are enabled automatically. In the left-side navigation pane, choose to view the results.
Quota consumption rules and pricing
In offline identification scenarios, the AI detection quota is consumed as follows:
Identifying 1 TB of text files requires approximately 100,000 text detection calls.
Identifying 10,000 data tables requires approximately 3,000,000 text detection calls.
Identifying 1 TB of images requires approximately 1,000,000 image detection calls.
In API call scenarios, the AI detection quota is consumed as follows:
Text detection — Each input text that contains at least one piece of sensitive information consumes one call. The length of a single input text cannot exceed 2,000 characters. Text that exceeds the limit is not detected.
Image detection — Each detected image consumes one call. A single image cannot exceed 10 MB. Images that exceed the limit are not detected.
When the AI detection quota is exhausted, DSC displays an insufficient quota message. Follow the on-screen instructions to upgrade.
Feature | Pricing |
AI Text Detection | Number of calls: 15 CNY/10,000 calls/month |
AI Image Detection | Number of calls: 30 CNY/10,000 calls/month |
Handling AI detection quota depletion
When your AI detection quota is depleted:
Text detection: All Identification Models stop invoking the AI LLM, reducing detection accuracy by 15% to 30%.
Image detection: DSC cannot detect images.
The DSC console prompts you to upgrade. Follow the on-screen instructions or click Upgrade on the Overview page. In the AI Data Security Detection section of the upgrade page, select the required number of AI Text Detection and AI Image Detection calls.
To conserve quota, you can manually control the AI LLM invocation status for Identification Models or OSS Identification Tasks.
Identification Models: Go to . On the Identification Models tab, find the target detection model and click the switch icon in the AI Model Invocation column to disable the feature. We recommend that you do not perform this operation unless necessary. For more information, see View and configure detection templates.
OSS Identification Tasks: Go to . On the Identification Tasks tab, select OSS and then click Create. On the Create page, turn off the AI-Powered Image Detection switch. For more information, see Scan for sensitive data using a detection task.
Appendix: Detection models that support LLM invocation
The following detection models support LLM invocation.
Address
Address (Malaysia)
Address (English)
Address (the Chinese mainland)
Residential Address
Name
Name (Malaysia)
Name (English)
Name (Traditional Chinese)
Name (Simplified Chinese)
Personal Name
Identity/Certificate
Passport Number (the Chinese mainland)
U.S. Social Security Number (SSN)
ID Card Number (Hong Kong (China))
Passport
ID Card
Contact Information
Landline Number (United States)
Landline Number (the Chinese mainland)
Personal Phone Number
Bank/Payment
Credit Card Number
Bank Card Number (the Chinese mainland)
Bank Account
Organization/Enterprise Qualification
Tax Registration Certificate Number
Unified Social Credit Code
Organization Code
Business License Number