Blog
Security research, vulnerability analysis, and product updates.
Allaccess-controlai-securityanchorannouncementsaptosauditbenchmarksbest-practicesbridgechecklistci-cdconfigurationcosmoscosmwasmdefidevopserc-3525evmexploit-teardownflash-loansgovernancelstmovemultisignorth-koreaoracleproductreentrancyresearchrustsecuritysocial-engineeringsolanasoliditysuisupply-chainsvmtestingtoolingupgradesvulnerability-class
|Odin Scan Team
How We Dramatically Improved OdinScan Accuracy
Same model, same benchmark, better results: a unified audit agent found all 17 ground-truth bugs on our hardest EVM contract with a single false positive, lifting F1 from 0.89 to 0.97.
researchbenchmarkssecurity
|Odin Scan Team
OdinBench v0.1: Measuring False Positives in LLM Security Scanners
Most AI security scanners ship without measuring themselves. We didn't. On 38 contracts across Solana and CosmWasm, our verification pipeline cuts false positives 45% while preserving recall, and on EVM every Critical-severity finding is a real bug.
researchbenchmarkssecurity