OpenAI and crypto investment firm Paradigm have joined forces to defend smart contracts from cyber attacks. The companies released a benchmark called EVMbench that evaluates AI agents' ability to detect, patch and exploit smart contract vulnerabilities, The Block reported on Feb. 18.
OpenAI said in a blog post, "Smart contracts secure billions of dollars of assets, and AI agents are a game-changing technology for both attackers and defenders." It said that as AI improves its ability to read, write and run code, it is important to measure that capability in economically meaningful environments and use it for defence.
EVMbench is built on 120 vulnerabilities drawn from 40 audits. It includes cases collected from a sponsorship-based open-source audit contest and from the security audit process for Tempo, a layer-1 blockchain jointly developed by Paradigm and Stripe. EVMbench offers evaluation modes covering vulnerability detection, contract fixes and removing exploitability. It also measures the ability to carry out an end-to-end fund theft attack against contracts deployed in a sandbox blockchain environment.
Chi-gyu Hwang
delight@d-today.co.kr