OpenAI and Paradigm Launch EVMbench to Test AI Agents Against Smart Contract Vulnerabilities

TLDR: EVMbench draws from 120 high-severity vulnerabilities curated across 40 real-world smart contract audits.  GPT-5.3-Codex scored 72.2% in exploit mode, far outperforming GPT-5, which only reached 31.9% in testing.  Moonwell and CrossCurve both suffered smart contract exploits recently, adding urgency to AI-driven security tools.  Anthropic’s late 2025 report warned that AI agents can already identify [...]

The post OpenAI and Paradigm Launch EVMbench to Test AI Agents Against Smart Contract Vulnerabilities appeared first on Blockonomi.