OpenAI and Paradigm launch EVMbench to explore the application of AI Agents in smart contract security

OpenAI and the crypto investment firm Paradigm announced a collaboration to launch EVMbench, aimed at assessing the capabilities of AI Agents in the field of smart contract security, including the identification, patching, and exploitation of high-risk vulnerabilities.

Both parties stated that as AI improves its capabilities in code writing and execution, it can be exploited by attackers but can also serve as a defense tool. Therefore, it is necessary to test the security capabilities of AI systems in environments with real economic significance. EVMbench will provide various testing modes, including vulnerability detection, contract modification, elimination of exploitability, and simulating complete attack processes in a sandbox chain environment.

At the time of the tool's launch, the DeFi protocol Moonwell and the cross-chain liquidity protocol CrossCurve recently encountered smart contract vulnerability attacks, raising concerns about AI-assisted code security. EVMbench is reportedly built on 120 selected vulnerability samples, covering multiple publicly audited and competition cases.

Industry analysts believe that as AI technology advances, the threshold for identifying smart contract vulnerabilities may further decrease, and the game between security defense and attack capabilities is accelerating.