A new research paper explores the effectiveness of large language models (LLMs) in detecting vulnerabilities within cross-chain bridges on the Solana blockchain. The study implemented a lock-and-mint bridge with four exploitable variants, including flaws inspired by the Cashio and Wormhole incidents. When provided with source code lacking comments, LLMs could diagnose vulnerabilities in only 24% of cases, often incorrectly flagging secure code. However, including original comments in the source code significantly improved diagnostic accuracy to 60%, highlighting the importance of code comments for LLM-based security reviews. AI
IMPACT LLMs can aid in identifying critical vulnerabilities in blockchain infrastructure, potentially improving security for cross-chain transactions.
RANK_REASON Academic paper detailing a novel application of LLMs for security analysis. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →