OpenAI and Paradigm Built a Benchmark for Hacking Smart Contracts
EVMbench tests how well AI agents can find, exploit, and fix high-severity vulnerabilities in blockchain smart contracts.
OpenAI just teamed up with crypto-native firm Paradigm to launch EVMbench — a new benchmark designed to measure how effectively AI agents can detect, exploit, and patch serious smart contract vulnerabilities.
The benchmark targets the Ethereum Virtual Machine ecosystem, where billions in value sit behind code that's often riddled with exploitable flaws. EVMbench evaluates AI across three distinct capabilities: spotting high-severity bugs, writing working exploits, and generating fixes.
It's a notable collision of two worlds. OpenAI brings the AI muscle. Paradigm brings deep blockchain expertise. Together they're essentially grading how good AI is at being both attacker and defender in smart contract security.
The goal is straightforward: make smart contracts safer by systematically benchmarking AI's ability to handle real blockchain security challenges. Whether that actually reduces the parade of DeFi hacks remains to be seen.