Morning Minute: OpenAI and Paradigm Turn Focus to Smart Contracts
OpenAI and Paradigm have launched EVMbench, a new open-source benchmarking tool that tests AI agents’ ability to detect, exploit, and patch critical vulnerabilities in Ethereum smart contracts. Drawing on 120 real-world, high-severity bugs from major audits—including Stripe’s payments blockchain Tempo—EVMbench measures AI performance in security tasks. Since the project started, state-of-the-art models’ bug detection and exploitation rates have risen from below 20% to over 70%. OpenAI also expanded the private beta of Aardvark, a dedicated security research agent, and committed $10 million in API credits for cybersecurity research. This collaboration highlights AI’s growing importance in defending (and potentially attacking) blockchain systems, as smart contract hacks have caused over $5 billion in losses to DeFi in two years. EVMbench both equips defenders and exposes new risks if used maliciously, signaling a deepening integration between advanced AI research and crypto security.
