OpenAI and Paradigm have joined forces to tackle the persistent threat of smart contract exploits. They have jointly released EVMbench, an open benchmarking framework designed to evaluate AI agents in detecting, patching, and exploiting vulnerabilities in Ethereum smart contracts.
This initiative leverages 120 high-severity vulnerabilities curated from real-world audits and security assessments. Initial results show significant progress, with top AI models now capable of exploiting over 70% of critical bugs, a substantial increase from less than 20% previously. OpenAI is further supporting defensive crypto research with $10 million in API credits through its Cybersecurity Grant Program and expanding its security research agent, Aardvark.
The development signals a maturing integration between AI and cryptocurrency. With over $5 billion lost to DeFi exploits in the past two years, EVMbench represents a critical step towards enhancing the security of decentralized finance protocols. The framework's ability to identify vulnerabilities at high accuracy underscores its potential as both a defensive tool and a preview of future AI-driven auditing.