OpenAI EVMbench

[OpenAI](https://startupintros.com/orgs/openai) and Paradigm launched EVMbench on February 18, 2026 — an open evaluation framework that benchmarks AI agents' ability to detect, patch, and exploit high-severity smart contract vulnerabilities.

The benchmark tests agents across three capabilities: vulnerability detection (finding bugs in Solidity/EVM contracts), patching (generating fixes), and exploitation (writing proof-of-concept exploits to verify severity). It's open-source, allowing any researcher or security team to evaluate AI models for smart contract auditing.

This is a notable collaboration between OpenAI (the world's leading AI company) and Paradigm (one of crypto's most influential research-driven investment firms). Smart contract hacks have cost the crypto industry over $10 billion since 2020. EVMbench creates the standard by which AI security tools will be measured — whoever builds the best-performing agent on this benchmark has a massive market opportunity in DeFi security.