LLMs show promise in repairing web application security flaws · 1 source tracked

Researchers have developed BAC-Bench, the first benchmark designed to evaluate how well Large Language Models (LLMs) can repair broken access control (BAC) vulnerabilities in web applications. This benchmark includes 46 real-world vulnerabilities from six web applications across four programming languages, each with a sandboxed environment and validation tools. Initial evaluations show that base LLMs can repair an average of 37.68% of these vulnerabilities, but with simple scaffolding techniques like Apply-Feedback and Navigate-and-Locate, this repair rate increases significantly, with the best LLM achieving 75.36% repair accuracy. AI

IMPACT Establishes a benchmark for LLM-based security vulnerability repair, potentially accelerating adoption in web application security.

RANK_REASON Academic paper introducing a new benchmark for LLM application in a specific security domain. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Apply-Feedback
• arXiv
• BAC-Bench
• Hugging Face
• LLMs
• Navigate-and-Locate
• web applications

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →