SOSecure: Safer Code Generation with RAG and StackOverflow Discussions
Fuente:
arXiv
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918364595617792 |
|---|---|
| author | Mukherjee, Manisha Hellendoorn, Vincent J. |
| author_facet | Mukherjee, Manisha Hellendoorn, Vincent J. |
| contents | Large Language Models (LLMs) are widely used for automated code generation. Their reliance on infrequently updated pretraining data leaves them unaware of newly discovered vulnerabilities and evolving security standards, making them prone to producing insecure code. In contrast, developer communities on Stack Overflow (SO) provide an ever-evolving repository of knowledge, where security vulnerabilities are actively discussed and addressed through collective expertise. These community-driven insights remain largely untapped by LLMs. This paper introduces SOSecure, a Retrieval-Augmented Generation (RAG) system that leverages the collective security expertise found in SO discussions to improve the security of LLM-generated code. We build a security-focused knowledge base by extracting SO answers and comments that explicitly identify vulnerabilities. Unlike common uses of RAG, SOSecure triggers after code has been generated to find discussions that identify flaws in similar code. These are used in a prompt to an LLM to consider revising the code. Evaluation across three datasets (SALLM, LLMSecEval, and LMSys) show that SOSecure achieves strong fix rates of 71.7%, 91.3%, and 96.7% respectively, compared to prompting GPT-4 without relevant discussions (49.1%, 56.5%, and 37.5%), and outperforms multiple other baselines. SOSecure operates as a language-agnostic complement to existing LLMs, without requiring retraining or fine-tuning, making it easy to deploy. Our results underscore the importance of maintaining active developer forums, which have dropped substantially in usage with LLM adoptions. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_13654 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | SOSecure: Safer Code Generation with RAG and StackOverflow Discussions Mukherjee, Manisha Hellendoorn, Vincent J. Software Engineering Cryptography and Security Large Language Models (LLMs) are widely used for automated code generation. Their reliance on infrequently updated pretraining data leaves them unaware of newly discovered vulnerabilities and evolving security standards, making them prone to producing insecure code. In contrast, developer communities on Stack Overflow (SO) provide an ever-evolving repository of knowledge, where security vulnerabilities are actively discussed and addressed through collective expertise. These community-driven insights remain largely untapped by LLMs. This paper introduces SOSecure, a Retrieval-Augmented Generation (RAG) system that leverages the collective security expertise found in SO discussions to improve the security of LLM-generated code. We build a security-focused knowledge base by extracting SO answers and comments that explicitly identify vulnerabilities. Unlike common uses of RAG, SOSecure triggers after code has been generated to find discussions that identify flaws in similar code. These are used in a prompt to an LLM to consider revising the code. Evaluation across three datasets (SALLM, LLMSecEval, and LMSys) show that SOSecure achieves strong fix rates of 71.7%, 91.3%, and 96.7% respectively, compared to prompting GPT-4 without relevant discussions (49.1%, 56.5%, and 37.5%), and outperforms multiple other baselines. SOSecure operates as a language-agnostic complement to existing LLMs, without requiring retraining or fine-tuning, making it easy to deploy. Our results underscore the importance of maintaining active developer forums, which have dropped substantially in usage with LLM adoptions. |
| title | SOSecure: Safer Code Generation with RAG and StackOverflow Discussions |
| topic | Software Engineering Cryptography and Security |
| url | https://arxiv.org/abs/2503.13654 |