OpenAI and Paradigm Launches EVMbench to Test AIs on Smart Contract Security (timescrypto.com)

OpenAI and Paradigm have introduced EVMbench, a new benchmarking tool designed to assess the capabilities of AI agents in identifying, patching, and exploiting vulnerabilities in Ethereum-based smart contracts, which secure over $100 billion in crypto assets. EVMbench focuses on 120 types of vulnerabilities derived from 40 security audits, simulating scenarios relevant to payment-oriented smart contracts. The evaluation process consists of three key metrics: detecting vulnerabilities through audits, patching contracts while preserving functionality, and exploiting vulnerabilities in a controlled environment. The framework employs an Anvil-based, Rust-language harness to ensure consistent and reproducible evaluations without risking live network assets. This initiative is significant for the AI/ML community as it advances the understanding of AI's effectiveness in cybersecurity, especially within the rapidly evolving crypto landscape. Recent tests showed that AI models, like the latest GPT-5.3-Codex, have dramatically improved performance on exploitation tasks compared to previous versions. While detection and patching remain challenging for AI agents, EVMbench not only addresses the evaluation of potential threats but also promotes defensive strategies in cybersecurity. Furthermore, OpenAI's commitment of $10 million in API credits through its Cybersecurity Grant Program underlines a broader effort to enhance research initiatives and bolster the development of robust security solutions.