A multi-agent based automated penetration testing method and apparatus
By combining a multi-agent architecture and a large language model, intelligent penetration testing is achieved, which solves the problems of low intelligence and high reliance on human labor in existing penetration testing technologies, improves vulnerability discovery rate and testing efficiency, and ensures the security and consistency of the testing process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING BAIGEFEICHI TECH LLC
- Filing Date
- 2026-04-01
- Publication Date
- 2026-07-17
AI Technical Summary
Existing penetration testing techniques rely on human experience, cannot intelligently plan attack paths, and are difficult to cover complex business logic vulnerabilities. Furthermore, automated tools lack context awareness and dynamic adjustment capabilities, resulting in high false negative rates, numerous false positives, high labor costs, and long testing cycles.
A multi-agent architecture is adopted, introducing a root agent and sub-agents to cooperate in a hierarchical manner. A large language model is used for policy generation and task decomposition. Through semantic analysis and closed-loop decision-making, real attack behaviors are simulated to achieve dynamic task orchestration and efficient execution.
It significantly reduces reliance on human experts, improves vulnerability discovery rates and testing efficiency, ensures contextual consistency and non-destructiveness, and adapts to the stable execution of complex business logic and long-chain tasks.
Smart Images
Figure CN122412290A_ABST