Security Center provides security detection capabilities for AI applications, such as application vulnerability detection, asset exposure analysis, configuration checks, and image risk scanning. These tools help you ensure the compliance and security of your AI applications. This topic describes the security capabilities that Security Center provides for AI asset inventory and AI application risk management.
Prerequisites
If you create an AI application on a server, you must install the Security Center agent on the server. For more information, see Install the agent.
When you purchase an Alibaba Cloud Elastic Compute Service (ECS) instance, the Security Center agent is automatically installed because the Free Security Hardening option is selected by default.
AI asset inventory
Security Center can automatically and continuously discover AI applications deployed in the following Alibaba Cloud services:
Elastic Compute Service (ECS): instances, images, and snapshots.
Platform for AI (PAI): container images and instances created on the PAI platform, such as Distribution Switch (DSW) instances for interactive modeling and online model service instances.
Intelligent Computing LINGJUN
View AI asset overview
In the AI Asset area on the page of the Security Center console, you can view the total number of AI assets and the number of assets with threats.
View AI components on servers
The Asset Fingerprints feature of Security Center collects information about AI components on servers. For more information, see Asset Fingerprints.
On the AI Component tab of the page in the Security Center console, you can view the AI components on your server assets, such as AI applications, AI tools, and AI services.
View AI assets by tag
View PAI assets on the Cloud Products page
Security Center synchronizes cloud product data from Alibaba Cloud Platform for AI (PAI) in real time.
On the page of the Security Center console, you can set Tag to AI Application in the search box to view the assets on the PAI platform and their risk status. For more information, see View cloud product information.
View AI assets on the Image Repositories page
After you enable the container image scan feature, Security Center automatically synchronizes image repositories from Alibaba Cloud Container Registry and detects container image assets created on the PAI platform.
In the Security Center console, go to the page. On the Repository Images tab, select to view the list of image repositories for Platform for AI (PAI) and their risk status. For more information, see View image information.
AI application risk management
View pending AI risks
On the Overview page of the Security Center console, Security Center provides statistics on pending AI security risks to help you view AI application risks from different dimensions. In the AI Security Risk panel, click Handle Now to navigate to the corresponding module to manage risks and reduce the attack surface.
You can also view the number of AI assets with threats in the AI Asset area on the Asset Overview page.
AI application vulnerabilities
Security Center detects and centrally displays vulnerabilities in various AI components, such as ollama and lm-studio. Vulnerabilities related to AI applications can lead to data breaches or unauthorized access to AI models and resources. Address these vulnerabilities promptly and manually.
In the Security Center console, go to the page and click the Application Vulnerability tab. Select Display Only AI-related Vulnerabilities to view vulnerabilities in AI applications. For more information, see View and manage vulnerabilities.
Internet-exposed AI components
Security Center can detect whether AI applications deployed on servers are exposed to the Internet. This helps you identify the attack surface of your AI assets on the Internet so that you can take timely measures to prevent attacks on your AI applications. For more information, see Asset exposure analysis.
On the page of the Security Center console, click the number in the AI Application Components area. The AI Application Components panel opens and lists the AI applications exposed to the Internet.
Security Center automatically adds the AI Application tag to assets that contain AI components. For an asset with the AI Application tag, click Exposure Details in the Actions column to view its details. Security Center marks AI components with a blue underline.
AI configuration risks
Security Center provides automated detection of security configuration risks for AI assets. The Cloud product configuration risk check feature is based on the AI security best practices of Alibaba Cloud, Azure, and AWS. This feature detects key security configuration items for AI application components, such as Platform for AI (PAI), AI services, and machine learning.
In the Security Center console, navigate to the page. On the Cloud Product Configuration Risks tab, click AI Configuration Management (AI-SPM) in the All Check Items area to view the related check items and their results. To remediate failed check items, see View and handle failed check items.
Image security scan
Security Center can scan container images created on Platform for AI (PAI) for risks, including image vulnerabilities, image baseline risks, malicious samples, and sensitive files. In addition to basic image security scanning, Security Center can detect the risk of storing AI service API call keys as plaintext in images, such as OpenAI-related keys and Alibaba Cloud PAI-EAS service tokens.
On the page in the Security Center console, you can view security risks in images. On the details page of a security risk, Security Center automatically applies the PAI tag to images that are created by PAI. For more information, see View and handle detected image risks.
Risks scanned by agentless detection
Security Center provides an agentless detection feature to comprehensively scan for security risks in Elastic Compute Service (ECS) instances that have AI components installed. These risks include vulnerabilities, baseline risks, malicious samples, and sensitive files. In addition to basic security risk scanning, Security Center can detect sensitive information related to AI API calls that is stored in plaintext on servers, such as OpenAI-related keys and Alibaba Cloud PAI-EAS service tokens. For more information, see Agentless detection.
On the page in the Security Center console, you can view the risk detection results for ECS instances.
