OpenAI and Paradigm don launch EVMbench to test AI on EVM smart‑contract security
OpenAI and crypto investor Paradigm don release EVMbench, na na open-source benchmark suite wey fit test AI agents and automated tools for Ethereum Virtual Machine (EVM) smart-contract security. Na follow-up to earlier announcement, full release pack 120 high-severity vulnerabilities wey dem collect from 40 audits (include public audit competitions and Tempo’s security work) and e come with labelled test cases, taxonomy of flaw types (reentrancy, access control, integer bugs, logic errors, and so on), plus reproducible evaluation tooling. EVMbench dey run agents for three modes — detect (find known flaws), patch (propose fixes without breaking functionality) and exploit (try controlled fund drains for isolated sandbox) — so teams fit measure detection rates, false positives, false negatives and coverage gaps across models and regular static-analysis scanners. Early results show say performance vary wide by task: newer models (especially OpenAI’s GPT-5.3-Codex in initial tests) do better for exploit tasks, while detection and patching still no perfect. OpenAI and Paradigm stress transparency: datasets, evaluation scripts and documentation na public so people fit reproduce comparisons and contribute. Project dey serve as both measurement tool and warning — as AI sabi improve, e fit help defenders and attackers alike — which mean say we need stronger defenses and more rigorous auditing. For crypto traders, EVMbench fit affect market risk indirectly over time by improving automated detection and patching of DeFi vulnerabilities, fit reduce exploit frequency and protocol risk, but immediate price effects no sure.
Neutral
EVMbench na tool for research an measurement wey dey target beta detection, patching and exploit simulation for EVM-based smart contracts. For traded assets wey dem mention or dey implied (Ethereum and DeFi tokens), news naa neutral for short term because publishing benchmark and early model results no go quick change exploit rates or token fundamentals. For medium to long term, if more people start to use better automated auditing, e fit reduce protocol-level risks and how often exploits happen, wey fit small positive for Ethereum-based DeFi tokens by lowering systemic security risk and boosting confidence for protocols. On the other hand, the benchmark show say AI fit still use to craft exploits, one factor wey fit keep or increase attack sophistication until defenses catch up. Traders suppose treat this as gradual, structural shift: small short-term price reaction, possible modest positive impact on risk premia for well-audited protocols over time, but make dem still dey cautious as AI-assisted offensive capabilities dey evolve.