OpenAI and crypto investment firm Paradigm announced a collaboration to launch EVMbench, a tool for evaluating the capabilities of AI agents in smart contract security, including identifying, patching, and exploiting high-risk vulnerabilities. The two companies stated that as AI's capabilities in code writing and execution improve, it can be exploited by attackers or used as a defensive tool. Therefore, it is necessary to test the security capabilities of AI systems in environments with real-world economic significance. EVMbench will offer multiple testing modes, including vulnerability detection, contract modification, exploit elimination, and simulating a complete attack process in a sandbox on-chain environment. This tool's launch comes as DeFi protocol Moonwell and cross-chain liquidity protocol CrossCurve have recently suffered smart contract vulnerability attacks, raising concerns about the security of AI-assisted code. EVMbench is reportedly built on 120 carefully selected vulnerability samples, covering multiple public audits and competition cases. Industry analysts believe that with advancements in AI technology, the threshold for identifying smart contract vulnerabilities may further decrease, and the interplay between security defenses and attack capabilities is accelerating.