OpenAI and Paradigm have partnered to launch EVMbench, an AI agent tool designed to improve smart contract security.

According to Mars Finance, market sources indicate that OpenAI and Paradigm have partnered to launch the EVMbench benchmark test, designed to evaluate the ability of AI agents to detect, patch, and exploit high-severity vulnerabilities in smart contracts, supporting end-to-end fund theft attacks in sandboxed blockchain environments. The tool is based on 120 carefully selected vulnerabilities (from 40 audits, including a security audit of Tempo, a Layer 1 blockchain co-developed by Paradigm and Stripe), aiming to encourage defensive applications to improve contract auditing and security. OpenAI stated in a blog post, "Smart contracts protect billions of dollars in assets, and AI agents will be revolutionary for both attackers and defenders." This move addresses recent vulnerability incidents such as the AI-assisted vulnerability in Moonwell and the $3 million cross-chain loss in CrossCurve, with Anthropic reporting that AI can reduce the cost of cryptographic attacks.