OpenAI and Paradigm Release EVMbench to Explore AI Agent Applications in Smart Contract Security

February 19, OpenAI and crypto investment firm Paradigm announced a partnership to launch EVMbench—an initiative aimed at evaluating AI agents’ capabilities in smart contract security, including identifying, patching, and exploiting critical vulnerabilities. The partners noted that as AI’s code generation and execution capabilities advance, it could be used by attackers while also acting as a defensive tool. Thus, testing AI systems’ security capabilities in an environment with real economic stakes is critical. EVMbench offers multiple testing modes: vulnerability detection, contract mutation, exploit eradication, and simulation of full attack flows in a sandboxed blockchain environment. The tool’s launch comes as DeFi protocol Moonwell and cross-chain liquidity platform CrossCurve recently faced smart contract vulnerability exploits—raising concerns about AI-assisted code security. EVMbench is built on 120 curated vulnerability samples, spanning public audits and competition cases. (Source: The Block)