/ common / websocket / static / aigdocs / docs / ai-infra-scan_en.md
ai-infra-scan_en.md
  1  # AI Infra Scan
  2  
  3  ## Feature Overview
  4  AI Infra Security Scan identifies known vulnerabilities (e.g., CVEs) in web services of AI infrastructure components  through precise fingerprint matching. This enables rapid detection of security gaps, empowering teams to mitigate risks proactively and maintain   secure, stable AI operations.
  5  
  6  ## Core Features
  7  - **Comprehensive Coverage**: Identifies **57 mainstream AI frameworks**, covering **1000+ known vulnerabilities** (CVEs).  
  8  - **Flexible Deployment**: Supports **single-target**, **batch**, and **local service ** scanning.  
  9  - **Intelligent Matching**: **YAML-based fingerprint rules** ensure high-precision detection accuracy.  
 10  - **Extensibility**: Enables **custom vulnerability templates** and **fingerprint rules** for specialized deployment scenarios.
 11  
 12  ## Quick Start
 13  
 14  ### WebUI Interface Workflow
 15  
 16  1.Select `AI Infra Scan` from the main page.
 17  
 18  2.Configure Scan Targets
 19     - Enter single/multiple URLs or IP addresses (one per line)
 20     - Import target lists via `.txt` file upload
 21     - ✨ *IP inputs trigger comprehensive port scanning*
 22        (Automatically checks common open ports)
 23  
 24  3.Select a MLLM to detect unauthenticated vulnerabilities, recommended GPT5/Gemini Pro/Sonnet4.5.
 25  
 26  4.Click `Send Message` button to initiate automated vulnerability detection.
 27     Results will populate in real-time upon completion.
 28  
 29  ![image-20250717185311173](./assets/image-20250717185311173-en.png)
 30  
 31  ![image-20250717185509861](./assets/image-20250717185509861-en.png)
 32  
 33  ## Fingerprint & Vulnerability Database
 34  
 35  ### Built-in Fingerprint Repository
 36  A.I.G includes **an extensive library of pre-configured AI component fingerprints**, accessible via the Plugin Management interface:
 37  
 38  
 39  1. **Access plugin management**
 40     Navigate to `Plugin Management` (bottom-left of the main page)
 41  2. **Review Built-in Resources**
 42     View all default fingerprint rules with search/filter capabilities
 43  3. **Manage Fingerprints**
 44     Perform real-time operations:
 45     - 🔍 Search rules by name/description/contributor
 46     - ➕ Add custom fingerprints and associated vulnerabilities
 47     - ✏️ Edit existing fingerprints and associated vulnerabilities
 48  
 49  ![image-20250814173036377](./assets/image-20250814173036377-en.png)
 50  ▶️ Changes apply immediately – subsequent scans automatically utilize updated databases
 51  ![image-20250717185223588](./assets/image-20250717185223588-en.png)
 52  
 53  ## Supported AI Components & Vulnerability Coverage
 54  
 55  A.I.G delivers comprehensive security coverage for critical AI infrastructure components. Current supported components and vulnerability statistics:
 56  
 57  | Category                   | Component Name          | Vulnerability Count | Risk Level  |
 58  | -------------------------- | ----------------------- | ------------------- | ----------- |
 59  | **Model Serving**          | gradio                  | 48                  | High        |
 60  |                            | ollama                  | 26                  | Medium-High |
 61  |                            | triton-inference-server | 33                  | Medium-High |
 62  |                            | vllm                    | 46                  | Medium      |
 63  |                            | xinference              | 3                   | Low         |
 64  |                            | fastchat                | 7                   | Medium      |
 65  |                            | llama-cpp               | 3                   | Low         |
 66  |                            | llmstudio               | 1                   | Low         |
 67  |                            | ChatRTX                 | 1                   | Low         |
 68  |                            | kubeai                  | 1                   | High        |
 69  | **LLM App Frameworks**     | langchain               | 51                  | High        |
 70  |                            | dify                    | 21                  | High        |
 71  |                            | anythingllm             | 17                  | Medium-High |
 72  |                            | open-webui              | 39                  | Medium-High |
 73  |                            | ragflow                 | 11                  | Medium      |
 74  |                            | qanything               | 8                   | Medium      |
 75  |                            | langflow                | 25                  | Medium      |
 76  |                            | litellm                 | 20                  | Medium      |
 77  |                            | mlflow                  | 66                  | High        |
 78  |                            | librechat               | 10                  | Medium      |
 79  |                            | nextchat                | 4                   | Medium      |
 80  |                            | lobechat                | 4                   | Medium      |
 81  |                            | lobehub                 | 1                   | Medium      |
 82  |                            | flowise                 | 3                   | Medium      |
 83  |                            | langfuse                | 3                   | Low         |
 84  |                            | new-api                 | 2                   | Medium      |
 85  |                            | Chuanhugpt              | 27                  | Medium-High |
 86  |                            | crewai                  | 3                   | Critical    |
 87  | **Data Processing**        | clickhouse              | 25                  | High        |
 88  |                            | feast                   | 2                   | Low         |
 89  |                            | dask                    | 3                   | Low         |
 90  | **Visualization & UI**     | jupyter-server          | 14                  | Medium-High |
 91  |                            | jupyterlab              | 8                   | Medium      |
 92  |                            | jupyter-notebook        | 2                   | Low         |
 93  | **Workflow Orchestration** | kubeflow                | 4                   | Medium      |
 94  |                            | ray                     | 10                  | Medium      |
 95  |                            | n8n                     | 33                  | Medium-High |
 96  |                            | simstudioai             | 9                   | Medium      |
 97  | **Other AI Components**    | comfyui                 | 8                   | Low         |
 98  |                            | comfy_mtb               | 1                   | Low         |
 99  |                            | ComfyUI-Prompt-Preview  | 1                   | Low         |
100  |                            | ComfyUI-Custom-Scripts  | 1                   | Low         |
101  |                            | ComfyUI-Impact-Pack     | 1                   | Low         |
102  |                            | ComfyUI-Manager         | 1                   | Low         |
103  |                            | ComfyUI-Ace-Nodes       | 1                   | Low         |
104  |                            | ComfyUI-Bmad-Nodes      | 1                   | Low         |
105  |                            | pyload-ng               | 25                  | Medium      |
106  |                            | kubepi                  | 5                   | Medium      |
107  |                            | llamafactory            | 4                   | Low         |
108  |                            | bentoml                 | 2                   | Medium      |
109  |                            | blinko                  | 1                   | Low         |
110  |                            | weknora                 | 2                   | Low         |
111  |                            | pinchtab                | 6                   | Medium-High |
112  |                            | wallos                  | 1                   | Low         |
113  |                            | praisonai               | 2                   | Medium      |
114  |                            | text-generation-webui   | 1                   | Medium      |
115  |                            | openclaw                | 566                 | Medium-High |
116  | **AI Agent Config Security** | AI-Agent-Config       | 1                   | High        |
117  | **Total**                  |                         | **1000+**           |             |
118  
119  > **Note**: The vulnerability database is continuously updated. Regular scanning of high-risk components is recommended.
120  
121  ## Fingerprint Matching Rule Details
122  
123  ### Rule Structure
124  
125  AI Infra Guard uses YAML format to define fingerprint matching rules, which mainly include the following parts:
126  
127  ```yaml
128  info:
129    name: Component Name
130    author: Rule Author
131    severity: Information Level
132    metadata:
133      product: Product Name
134      vendor: Vendor Name
135  http:
136    - method: HTTP Request Method
137      path: Request Path
138      matchers:
139        - Matching Conditions
140  ```
141  
142  ### Example: Dify Fingerprint Rule
143  
144  ```yaml
145  info:
146    name: dify
147    author: Tencent Zhuque Lab
148    severity: info
149    metadata:
150      product: dify
151      vendor: dify
152  http:
153    - method: GET
154      path: '/'
155      matchers:
156        - body="<title>Dify</title>" || icon="97378986"
157  version:
158    - method: GET
159      path: '/console/api/version'
160      extractor:
161        part: header
162        group: 1
163        regex: 'x-version:\s*(\d+\.\d+\.?\d+?)'
164  ```
165  
166  ### Matcher Syntax Explanation
167  
168  #### Match Locations
169  
170  | Location | Description             | Example                                   |
171  | -------- | ----------------------- | ----------------------------------------- |
172  | `title`  | HTML page title         | `title="Gradio"`                          |
173  | `body`   | HTTP response body      | `body="gradio-config"`                    |
174  | `header` | HTTP response header    | `header="X-Gradio-Version: 3.34.0"`       |
175  | `icon`   | Website favicon hash    | `icon="d41d8cd98f00b204e9800998ecf8427e"` |
176  
177  #### Logical Operators
178  
179  | Operator | Description                               | Example                                                      |
180  | -------- | ----------------------------------------- | ------------------------------------------------------------ |
181  | `=`      | Fuzzy contains match (case-insensitive)   | `body="gradio"`                                              |
182  | `==`     | Exact equals match (case-sensitive)       | `header="Server: Gradio"`                                    |
183  | `!=`     | Not equals match                          | `header!="Server: Apache"`                                   |
184  | `~=`     | Regular expression match                  | `body~="Gradio v[0-9]+.[0-9]+.[0-9]+"`                       |
185  | `&&`     | Logical AND                               | `body="gradio" && header="X-Gradio-Version"`                 |
186  | `||`     | Logical OR                                | `body="gradio" || body="Gradio"`                             |
187  | `()`     | Grouping to change precedence             | `(body="gradio" || body="Gradio") && header="X-Gradio-Version"` |
188  
189  ## Operational Best Practices
190  
191  
192  1.**Schedule Regular Scans**: Schedule weekly comprehensive scans of your AI infrastructure to promptly identify emerging vulnerabilities.
193  
194  2.**Prioritize High-Risk Components**: Focus scanning resources on components with high vulnerability densities, such as Gradio, LangChain, and ClickHouse.
195  
196  3.**Extend with Custom Rules**: Enhance detection capabilities for organization-specific AI components by adding custom fingerprint rules.
197  
198  4.**Integrate into CI/CD Pipelines**: Embed security scanning into the continuous integration (CI) process for AI applications to implement shift-left security.
199  
200  5.**Track Vulnerability Remediation**: Establish a tracking mechanism for vulnerabilities discovered during scans to ensure timely remediation.
201  
202  By leveraging the AI Infra Scan service, you can effectively identify potential security risks within your AI systems, providing robust assurance for building a secure and reliable AI infrastructure.