Efficient audits with machine learning and Slither-simil

Summary: Trail of Bits built Slither-simil, a machine learning addition to its Slither static analyzer that encodes Solidity functions as SlithIR tokens and scores structural similarity against a database of vulnerabilities pulled from prior client audits. The experiment drew on 89 vulnerabilities found across 10 security assessments, targeting flaws Slither itself could not detect, and the team cites data scarcity and the effort of compiling projects across different solc versions as the main bottlenecks.
What's compelling: The tool is trained on 89 vulnerabilities from 10 audits, so its usefulness to any new client depends on how closely that small, proprietary corpus resembles their code - a limit the team itself flags as data scarcity in a young language.

Read the original at Trail of Bits Blog ↗