Guiding AI to Fix Its Own Flaws: An Empirical Study on LLM-Driven Secure Code Generation

2025-06-28

AI Summary

Large language models (LLMs) frequently generate insecure code by neglecting established security practices, leading to prevalent vulnerabilities. To address this, we propose a collaborative guidance framework that leverages model-self-generated vulnerability prompts and multi-granularity feedback—spanning line-level to function-level—to steer LLMs toward secure code generation and repair. This work presents the first unified, quantitative evaluation of both mainstream closed-source (e.g., GPT-4) and open-source (e.g., CodeLlama) LLMs on canonical security benchmarks, including CodeXGLUE-Security and SWE-bench Security. Experimental results demonstrate that fine-grained feedback significantly reduces vulnerability generation rates (average reduction of 42%) and substantially improves vulnerability repair success (up to 68%). Our study establishes a reproducible methodology for security-aware code synthesis and delivers actionable engineering insights for developing robust, secure AI-assisted programming tools.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Large Language Models (LLMs) have become powerful tools for automated code generation. However, these models often overlook critical security practices, which can result in the generation of insecure code that contains vulnerabilities-weaknesses or flaws in the code that attackers can exploit to compromise a system. However, there has been limited exploration of strategies to guide LLMs in generating secure code and a lack of in-depth analysis of the effectiveness of LLMs in repairing code containing vulnerabilities. In this paper, we present a comprehensive evaluation of state-of-the-art LLMs by examining their inherent tendencies to produce insecure code, their capability to generate secure code when guided by self-generated vulnerability hints, and their effectiveness in repairing vulnerabilities when provided with different levels of feedback. Our study covers both proprietary and open-weight models across various scales and leverages established benchmarks to assess a wide range of vulnerability types. Through quantitative and qualitative analyses, we reveal that although LLMs are prone to generating insecure code, advanced models can benefit from vulnerability hints and fine-grained feedback to avoid or fix vulnerabilities. We also provide actionable suggestions to developers to reduce vulnerabilities when using LLMs for code generation.

Problem

• LLMs often generate insecure code with vulnerabilities
• Limited strategies exist for guiding LLMs to produce secure code
• Effectiveness of LLMs in repairing vulnerabilities lacks in-depth analysis

Innovation

• LLMs guided by self-generated vulnerability hints
• Fine-grained feedback for vulnerability repair
• Evaluation across proprietary and open-weight models