Scribd
Upload
268 views14 pages

Pentestgpt PPT

Uploaded by

victomteng
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
268 views14 pages

Pentestgpt PPT

Uploaded by

victomteng
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 14

PentestGPT: Automated

Penetration Testing

Gelei Deng
Nanyang Technological University
01
Security
Testing
Security Testing
• The process of examining system security with
testing
• Approaches
• White-box Testing – Audit
• Accurate and efficient
• Privacy concerns
• Black-box Testing – Penetration Testing
• Typically requires more domain knowledge
• Hard to automate
Penetration Testing
• The Most Widely Used Blackbox Approach
• General Procedure
1. Test the system with tools
2. Interpret testing results
3. Check Exploitability
• Develop the exploit, or
• Go to step 1
Penetration Testing
• Key Factor
• The tester’s domain knowledge
• How to use the tools? How to
interpret the test results?
• Drawbacks
• Not reproducible
• Rely on penetration tester
Penetration Testing
• Penetration Testing is a manual effort
• Automated tools for sub-steps
• Testers need to decide on the next step manually
• Existing automation approaches are not end-to-end automatic
• Scanners are not end-to-end
• Machine learning-based approaches (Reinforcement learning, etc.) are not mature

Is it possible to automate penetration testing process?


02
LLM for
Pentest
LLM Assisted Pentesting
• Use LLM as the reasoning module
• Human tester without domain knowledge to perform the test
• Execute the command generated by LLM
• Return the outputs back to LLM
Pros & Cons of LLM
• Pros
• LLMs have knowledge on penetration testing
• LLMs can generate commands to use tools
• LLMs have understanding on commonly seen service.

• Cons
• Token Limit: Context cannot be maintained
• Attention Issue: LLM focus on most recent commands
• Hallucination: generate incorrect/non-related information as conversation
continues
03
Pentest
GPT
Penetration Testing
• We propose PentestGPT, an LLM-powered penetration
testing solution
• An open-source project with more than 6.5k stars on GitHub
• Arxiv Paper, to be presented at USENIX Security 24’
• Demo Video
• PentestGPT can
• Perform end-to-end automatic penetration testing without
human guidance
• Exploit the vulnerabilities to gain system access
• Automatically generate penetration testing report Arxiv Preprint
PentestGPT
• Our LLM-powered penetration testing solution
• Optimized to counter the drawbacks of LLMs
• Token Length Limit – Parsing Module
• Reasoning Capability – Reasoning Module
• Hallucination – Generation Module
PentestGPT Demo
THANKS

You might also like