Scribd
Upload
5 views7 pages

Research Topic

The document discusses the distinction between machine auditable and unauditable bugs in vulnerability detection, highlighting examples of each type. It identifies research gaps in current tools, particularly their limitations in detecting exploitable bugs and providing explanations. The proposed methodology involves using graph-based approaches to analyze smart contracts and improve vulnerability detection through advanced techniques like Node2Vec and LLM integration.

Uploaded by

Kiên Lê
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
5 views7 pages

Research Topic

The document discusses the distinction between machine auditable and unauditable bugs in vulnerability detection, highlighting examples of each type. It identifies research gaps in current tools, particularly their limitations in detecting exploitable bugs and providing explanations. The proposed methodology involves using graph-based approaches to analyze smart contracts and improve vulnerability detection through advanced techniques like Node2Vec and LLM integration.

Uploaded by

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

Graph-based RAG for Vulnerability Detection

Research Progress
Machine auditable bugs
• Bugs that can be detected automatically using static or dynamic
analysis tools
▪ Pattern-based detection
▪ No complex business logic reasoning required
• Examples
▪ Reentrancy
▪ Integer Overflow and Underflow
▪ Mishandled Exception
▪ Unprotected SELFDESTRUCT
Machine unauditable bugs
• Bugs that require human reasoning and understanding of
business logic.
▪ Complex control flow
▪ Cross-function dependencies
▪ Business logic vulnerabilities
• Examples
▪ Price oracle manipulation
▪ Access Control Vulnerability
▪ Business Logic Vulnerability
▪ Atomicity Violation
Relevant papers
• Review for Vulnerability [Link]
Research gaps
• Most tools detect common bugs but struggle with exploitable
bugs (e.g., price oracle manipulation)
▪ Oyente, ReDefender, CONFUZZIUS, ByteEye, PropertyGPT, SCALM
• Some tools target exploitable bugs but only use a small dataset
for evaluation
▪ DeepDesc (210 contracts are affected by exploitable bugs)
• Some tools could detect bugs in smart contracts but don’t provide
the explanation
▪ Oyente, ReDefender, CONFUZZIUS, ByteEye, PropertyGPT, DeepDesc,
SmartInv
Research questions
• To what extent can graph-based or context-aware approaches
detect complex exploit scenarios in smart contracts?
• How does the performance of vulnerability detection tools change
when evaluated on large-scale real-world smart contract
datasets?
• How can detection tools generate meaningful explanations for
detected smart contract vulnerabilities?
Our proposed methodology
• Collect the dataset including machine unaditable bugs
• Convert their call graphs into embeddings using Node2Vec,
Graph2Vec, .etc
• Compute the similarity between malicious call graphs and the call
graph of the given contract
• Using RAG to get relevant call graphs and them fuse them and the
given contract into the input of a LLM ⇒ Use the LLM to find the
type of vulnerability and lines of code containing vulnerabilities

You might also like