OpenAI and Paradigm Launch Initiative to Protect Ethereum Smart Contracts Against AI Attacks

OpenAI has announced a partnership with crypto venture capital firm Paradigm to enhance the security of smart contracts on Ethereum. The collaboration responds to the growing threat posed by AI agents that are increasingly capable of analyzing code and exploiting vulnerabilities ^[1].

At the heart of the initiative is EVMbench, a new benchmark platform for evaluating AI agents in the context of Ethereum-compatible smart contracts. The platform not only enables the detection of critical security vulnerabilities, but also their targeted remediation as well as the simulation of complete attack scenarios in an isolated testing environment ^[1].

The database comprises 120 selected vulnerabilities from 40 audits, including insights from the development of Tempo, a Layer-1 blockchain developed by Paradigm and Stripe ^[1].

The timing of the initiative is significant: recently, several DeFi protocols were compromised, including Moonwell through AI-generated code and CrossCurve with damages of approximately three million US dollars. Experts warn that advanced AI agents could make exploits cheaper and therefore more frequent in the future ^[1].