OpenAI Ships EVMbench, a Benchmark for How Well AI Agents Can Hack — and Fix — Smart Contracts
OpenAI released a new benchmark measuring AI agents' ability to detect, exploit, and patch high-severity smart contract vulnerabilities — a move that simultaneously advances DeFi security and raises uncomfortable questions about dual-use capability.
OpenAI quietly released EVMbench on Wednesday — a benchmark designed to evaluate how well AI agents can find, exploit, and repair critical vulnerabilities in Ethereum Virtual Machine smart contracts. As @OpenAI described it, the benchmark measures performance across the full lifecycle of a vulnerability: detection, proof-of-concept exploitation, and patch generation.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.