Location
https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php
Document Type
Event
Start Date
20-11-2026 2:00 PM
Description
Finding a vulnerability starts the job. A verified repair completes it. We developed and tested an autonomous penetration testing and remediation agent that connects offensive security testing with defensive repair. The system brings vulnerability discovery, controlled validation, code changes, and verification into one workflow. A Java client coordinates a locally hosted language model and a custom Model Context Protocol (MCP) server. The model interprets evidence and selects the next action, while the server validates requests and executes approved tools. Nmap identifies exposed ports and services, while Nikto and curl support focused web checks and vulnerability confirmation. Remediation tools apply controlled code, configuration, or dependency changes. Target and file path restrictions, execution limits, backups, and audit logs constrain and document the agent's actions. Implementation and testing used an intentionally vulnerable Flask application with mock data inside an isolated Docker environment. Three predefined scenarios cover exposed configuration data, SQL injection in a login endpoint, and an outdated dependency. Evaluation compares evidence collected before and after remediation. A repair is considered successful only when the original vulnerability test no longer succeeds and legitimate application behavior remains functional. Reports connect findings, changes, and verification outcomes for review. The completed system demonstrates an integrated approach to AI directed security testing and remediation within a defined lab scope. Its contribution is a repeatable workflow that moves beyond identifying weaknesses to applying changes and evaluating their effects. Future work includes broader vulnerability coverage and evaluation on additional applications.
Included in
1348-UC PatchPilot: Autonomous Pentesting and Remediation
https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php
Finding a vulnerability starts the job. A verified repair completes it. We developed and tested an autonomous penetration testing and remediation agent that connects offensive security testing with defensive repair. The system brings vulnerability discovery, controlled validation, code changes, and verification into one workflow. A Java client coordinates a locally hosted language model and a custom Model Context Protocol (MCP) server. The model interprets evidence and selects the next action, while the server validates requests and executes approved tools. Nmap identifies exposed ports and services, while Nikto and curl support focused web checks and vulnerability confirmation. Remediation tools apply controlled code, configuration, or dependency changes. Target and file path restrictions, execution limits, backups, and audit logs constrain and document the agent's actions. Implementation and testing used an intentionally vulnerable Flask application with mock data inside an isolated Docker environment. Three predefined scenarios cover exposed configuration data, SQL injection in a login endpoint, and an outdated dependency. Evaluation compares evidence collected before and after remediation. A repair is considered successful only when the original vulnerability test no longer succeeds and legitimate application behavior remains functional. Reports connect findings, changes, and verification outcomes for review. The completed system demonstrates an integrated approach to AI directed security testing and remediation within a defined lab scope. Its contribution is a repeatable workflow that moves beyond identifying weaknesses to applying changes and evaluating their effects. Future work includes broader vulnerability coverage and evaluation on additional applications.