Sam Altman’s OpenAI Partners With Paradigm To Launch AI Tool EVMbench To Test Ethereum Security

ChatGPT creator OpenAI and crypto investment firm Paradigm have partnered for a new AI tool that will assess if it can fix Ethereum’s security woes.

On 18 February 2026, EVMbench was released, and it now tests ground to see how well AI agents can find and patch bugs in smart contracts. With billions previously lost to hacks, this initiative could be a massive step forward for DeFi safety.

So, what is EVMbench? It is an open-source benchmarking framework designed to rigorously test how well AI agents can analyze and interact with smart contracts on the Ethereum Virtual Machine (EVM). As AI models become more capable of reading and writing code, this tool measures their ability to act as both security auditors and potential attackers.

In a blogpost today, OpenAI said, “Together with Paradigm⁠, we’re introducing EVMbench, a benchmark evaluating the ability of AI agents to detect, patch, and exploit high-severity smart contract vulnerabilities. EVMbench draws on 120 curated vulnerabilities from 40 audits, with most sourced from open code audit competitions.”

new collab from @paradigm and @OpenAI:

evmbench is a benchmark and agent harness for exploiting smart contract bugs

a few months ago, the best models found <20% of critical, fund-draining @Code4rena bugs in our benchmark. today they find > 70% https://t.co/soOrCR38eO pic.twitter.com/2lr0WUVo2Q

— Alpin Yukseloglu (@0xalpo) February 18, 2026

DISCOVER: Best New Cryptocurrencies to Invest in 2026

As Giants Like BlackRock Expand Into Ethereum’s Staking Ecosystem, EVMbench Becomes Critical

Currently, humans review smart contracts to catch mistakes. EVMbench is critical because smart contracts secure over $100 billion in assets. As major financial players enter the space and institutions like BlackRock interact with Ethereum staking security, the stakes are higher than ever.

With EVMbench, the goal is to see if AI can act as a hyper-fast security guard that never sleeps, spotting “vending machine” flaws before thieves do.

EVMbench puts AI models through a rigorous boot camp designed to mimic real-world dangers.

According to OpenAI blog, newer models like GPT-5.3-Codex are getting surprisingly good at the “exploit” part, solving over 70% of critical bugs during testing. While fixing the code remains a challenge, this “gym” for AI helps developers improve defensive tools. It creates a foundation for stronger infrastructure, similar to the engineering efforts seen in the recent MegaETH mainnet launch.

DISCOVER: Top 20 Crypto to Buy in 2026

“Smart contracts routinely secure $100B+ in open-source crypto assets”

“As AI agents improve at reading, writing, and executing code, it becomes increasingly important to measure their capabilities in economically meaningful environments, and to encourage the use of AI systems defensively to audit and strengthen deployed contracts,” said OPenAI.

However, OpenAI added, “Our grading system is robust but imperfect.”