Skip to content
InfoResearchPreprintLLM-specific

LLMs as Validators for Cross-Chain Bridges: Evaluating Source-Level and Transaction-Level Attack Detection on Solana

Published
Record updated
View JSON

Summary

This study evaluates whether open-weight LLMs can diagnose vulnerabilities in a native Rust lock-and-mint cross-chain bridge on a local Solana validator, using four exploitable variants modeled on the Cashio and Wormhole incidents and Solana audit classes. Across 16,800 responses and 7,200 graded answers, models diagnosed 24% of vulnerable source inputs without comments while often flagging the secure bridge, and diagnosed 6% of non-signer transaction scenarios on average. Original comments, some describing the flaw, raised source diagnosis to 60%.