tools-only AI Infra Scan
Feature Overview
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.
Core Features
- Comprehensive Coverage: Identifies 36 mainstream AI frameworks, covering 404 known vulnerabilities (CVEs).
- Flexible Deployment: Supports single-target, batch, and **local service ** scanning.
- Intelligent Matching: YAML-based fingerprint rules ensure high-precision detection accuracy.
- Extensibility: Enables custom vulnerability templates and fingerprint rules for specialized deployment scenarios.
Quick Start
WebUI Interface Workflow
1.Select AI Infra Scan from the main page.
2.Configure Scan Targets
- Enter single/multiple URLs or IP addresses (one per line)
- Import target lists via
.txtfile upload - ✨ IP inputs trigger comprehensive port scanning (Automatically checks common open ports)
3.Select a MLLM to detect unauthenticated vulnerabilities, recommended GPT5/Gemini Pro/Sonnet4.5.
4.Click Send Message button to initiate automated vulnerability detection.
Results will populate in real-time upon completion.
Fingerprint & Vulnerability Database
Built-in Fingerprint Repository
A.I.G includes an extensive library of pre-configured AI component fingerprints, accessible via the Plugin Management interface:
- Access plugin management
Navigate to
Plugin Management(bottom-left of the main page) - Review Built-in Resources View all default fingerprint rules with search/filter capabilities
- Manage Fingerprints
Perform real-time operations:
- 🔍 Search rules by name/description/contributor
- ➕ Add custom fingerprints and associated vulnerabilities
- ✏️ Edit existing fingerprints and associated vulnerabilities
▶️ Changes apply immediately – subsequent scans automatically utilize updated databases
Supported AI Components & Vulnerability Coverage
A.I.G delivers comprehensive security coverage for critical AI infrastructure components. Current supported components and vulnerability statistics:
| Category | Component Name | Vulnerability Count | Risk Level |
|---|---|---|---|
| Model Serving | gradio | 42 | High |
| ollama | 7 | Medium-High | |
| triton-inference-server | 7 | Medium-High | |
| vllm | 4 | Medium | |
| xinference | 0 | Low | |
| LLM App Frameworks | langchain | 33 | High |
| dify | 11 | High | |
| anythingllm | 8 | Medium-High | |
| open-webui | 8 | Medium-High | |
| ragflow | 2 | Medium | |
| qanything | 2 | Medium | |
| Data Processing | clickhouse | 22 | High |
| feast | 0 | Low | |
| Visualization & UI | jupyter-server | 13 | Medium-High |
| jupyterlab | 6 | Medium | |
| jupyter-notebook | 1 | Low | |
| tensorboard | 0 | Low | |
| Workflow Orchestration | kubeflow | 4 | Medium |
| ray | 4 | Medium | |
| Other AI Components | comfyui | 1 | Low |
| comfy_mtb | 1 | Low | |
| ComfyUI-Prompt-Preview | 1 | Low | |
| ComfyUI-Custom-Scripts | 1 | Low | |
| pyload-ng | 18 | Medium | |
| kubepi | 5 | Medium | |
| llamafactory | 1 | Low | |
| Total | 200+ |
Note: The vulnerability database is continuously updated. Regular scanning of high-risk components is recommended.
Fingerprint Matching Rule Details
Rule Structure
AI Infra Guard uses YAML format to define fingerprint matching rules, which mainly include the following parts:
info:
name: Component Name
author: Rule Author
severity: Information Level
metadata:
product: Product Name
vendor: Vendor Name
http:
- method: HTTP Request Method
path: Request Path
matchers:
- Matching Conditions
Example: Gradio Fingerprint Rule
info:
name: dify
author: Tencent Zhuque Lab
severity: info
metadata:
product: dify
vendor: dify
http:
- method: GET
path: '/'
matchers:
- body="<title>Dify</title>" || icon="97378986"
version:
- method: GET
path: '/console/api/version'
extractor:
part: header
group: 1
regex: 'x-version:\s*(\d+\.\d+\.?\d+?)'
Matcher Syntax Explanation
Match Locations
| Location | Description | Example |
|---|---|---|
title |
HTML page title | title="Gradio" |
body |
HTTP response body | body="gradio-config" |
header |
HTTP response header | header="X-Gradio-Version: 3.34.0" |
icon |
Website favicon hash | icon="d41d8cd98f00b204e9800998ecf8427e" |
Logical Operators
| Operator | Description | Example |
|---|---|---|
= |
Fuzzy contains match (case-insensitive) | body="gradio" |
== |
Exact equals match (case-sensitive) | header="Server: Gradio" |
!= |
Not equals match | header!="Server: Apache" |
~= |
Regular expression match | body~="Gradio v[0-9]+.[0-9]+.[0-9]+" |
&& |
Logical AND | body="gradio" && header="X-Gradio-Version" |
| ` | ` | |
() |
Grouping to change precedence | `(body="gradio" |
Operational Best Practices
1.Schedule Regular Scans: Schedule weekly comprehensive scans of your AI infrastructure to promptly identify emerging vulnerabilities.
2.Prioritize High-Risk Components: Focus scanning resources on components with high vulnerability densities, such as Gradio, LangChain, and ClickHouse.
3.Extend with Custom Rules: Enhance detection capabilities for organization-specific AI components by adding custom fingerprint rules.
4.Integrate into CI/CD Pipelines: Embed security scanning into the continuous integration (CI) process for AI applications to implement shift-left security.
5.Track Vulnerability Remediation: Establish a tracking mechanism for vulnerabilities discovered during scans to ensure timely remediation.
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.