"The framework embeds the relevant code snippet, the data flow path and surrounding contextual information into a structured JSON prompt for a fine-tuned LLM. We fine-tuned Llama 3 8B on a high-quality dataset of vetted false positives and true vulnerabilities, specifically covering major flaw categories like those in the OWASP Top 10 to form the core of the Intelligent Triage layer."
"The following empirical results validate our hybrid approach. Our test dataset had 25 diverse open source projects based on their active development and language diversity (Python, Java, JavaScript), with 170 vulnerabilities as ground truth, sourced from public exploit databases and manual expert verification. Precision: In our implementation, we found the precision jumped to 89.5%. This is a massive leap not only over Semgrep's baseline of 35.7%, but also over a purely LLM-based approach (GPT-4), which achieved 65.5%."
Stage 1 runs the Semgrep SAST engine to identify potential security risks and extract intermediate representations such as data flow paths from source to sink. Stage 2 uses a fine-tuned Llama 3 8B model to perform intelligent triage by embedding code snippets, data flow paths, and contextual information into structured JSON prompts and asking focused exploitability questions. The fine-tuned model distinguishes true vulnerabilities from false positives by leveraging vetted datasets covering OWASP Top 10 flaw categories. Tests on 25 open-source projects with 170 ground-truth vulnerabilities show precision increased to 89.5% and false positives reduced from 225 to 20.
Read at InfoWorld
Unable to calculate read time
Collection
[
|
...
]