Research

Research & Publications

Studies on on-chain data, smart-contract behavior, AI-driven analysis and risk in financial protocols. Data and method are linked where we can share them.

01BENCHMARKMay 20268 minThe Repeated-Runs Effect: Why One Tool Audit Pass Catches Just Half the BugsWe ran six LLM-based audit tools five times each across five public Web3 contests — 145 runs in total. A single run captures 56% of what a tool can find on average. Two runs jump to 77%. Three runs reach 89%. Diminishing returns after that. The data, the per-tool curves, and what stochasticity means for anyone building on top of LLM audit agents.BENCHMARKMETHODOLOGYAIRead article

02AIMar 202645 min84 Tools, 60 Papers, One Question: Is AI Auditing Ready?A comprehensive analysis of the AI-driven smart contract auditing landscape. 84 tools, 16 benchmarks, 60+ academic papers. Independent recall at 30-40%, exploit rate at 72.2%, ~80% of bugs beyond AI reach.AIBENCHMARKINTEGRATIONRead article

03PDFMar 2026PDFReducing False Positives in Vulnerability Detection: A Survey of LLM-Based and Agentic ApproachesSecurity teams face overwhelming false positive rates: traditional SAST tools produce 68-95% false positives, SOC analysts routinely leave the majority of alerts uninvestigated, and analyst burnout is pervasive. This survey analyzes 100 sources (academic papers, vendor data, production case studies) covering 10 distinct AI/LLM-based approaches to false positive reduction in security tooling. We find that no single approach dominates: hybrid SAST+LLM pipelines eliminate 94-98% of false positives in one industrial study; triage memories offer fast ROI; and proof-of-concept validation achieves zero false positives by confirming exploitability. We categorize approaches along three axes, compare cost-effectiveness, and identify seven open problems.AIFALSE POSITIVESSURVEYDOI10.5281/zenodo.20009438View publication