Zongwei Li

A staggering 30% of popular NFT collections contain exploitable permission control vulnerabilities, leaving them open to unauthorized access and financial loss.

Existing smart contract vulnerability detectors are getting schooled: PSR\textsuperscript{2} achieves a whopping 94.69% F1-score in detecting atomicity violations, nearly doubling the performance of current tools.

Automated repair of smart contract vulnerabilities is now significantly more effective, achieving an 81.5% repair rate by combining LLMs with a knowledge graph of Ethereum contracts.

LLMs, when combined with rule-based analysis and iterative refinement, can significantly outperform existing tools in detecting subtle smart contract vulnerabilities arising from library misuse.

Navigate massive codebases with LLMs using a "scouting-first" approach that slashes token consumption without sacrificing reasoning accuracy.