InfoResearchPreprintLLM-specific
Found, but Not Fixed : From Software Vulnerability Discovery to Fixes with Iterative LLM Pipelines
- Published
- Record updated
Summary
The paper investigates whether an iterative, LLM-based pipeline can automatically repair vulnerable functions in real-world code, since code repair remains a bottleneck after vulnerability detection. The authors propose a generalized framework that uses a multi-stage feedback loop and benchmark it against GPT-4o, GPT-5.1, and Gemini 3 Flash, reporting that such a pipeline can achieve high-quality code repair. The evaluation framework, fine-tuned weights, and experimental dataset are provided on the authors' website.
Related items
- InfoProton’s Lumo, a privacy-focused alternative to ChatGPT and Claude, arrives on desktopSame vendor · TechCrunch (Security)
- MediumThe risks of device code authentication for AI coding agentsSame vendor · Microsoft Security Blog
- InfoCalifornia sues OpenAI, alleges 80-plus agent escapes hidden from publicSame vendor · CSO Online
- InfoFormal Verification of the Signal Protocol Using AISame vendor · Schneier on Security
- InfoCalifornia Sues OpenAI, Seeking Court-Ordered Changes After the Hugging Face HackSame vendor · SecurityWeek