OpenAI and Paradigm Launch EVMbench to Enhance Ethereum Security
- EVMbench tests AI agents on detecting, patching, and exploiting vulnerabilities in Ethereum smart contracts.
- The tool evaluates AI performance across three modes using curated vulnerabilities from open audit competitions.
- GPT-5.3-Codex achieved a success rate of 72.2% in exploit mode testing, outperforming GPT-5’s previous score of 31.9%.
- Weekly deployment of smart contracts on Ethereum reached an all-time high of 1.7 million in November, with recent weekly deployments at over half a million.
OpenAI and Paradigm’s EVMbench aims to enhance the security of Ethereum smart contracts by benchmarking AI agents’ ability to handle real-world vulnerabilities effectively. The tool uses scenarios from security audits to test detection, patching, and exploitation capabilities.
The development of EVMbench highlights the importance of measuring AI performance in economically significant environments as smart contract deployments increase rapidly on the Ethereum network. (Source)
