ChatGPT and Paradigm Stand Guard for Crypto: Introducing EVMbench for Protecting Smart Contracts

In the world of cryptocurrencies, where smart contracts manage assets worth over $100 billion, security remains one of the key issues. Yesterday, February 18, 2026, OpenAI, in partnership with the leading crypto venture firm Paradigm, introduced a revolutionary tool, EVMbench. This benchmark is designed to evaluate how well AI agents can detect, fix, and even exploit vulnerabilities in smart contracts based on the Ethereum Virtual Machine (EVM).

What is EVMbench and Why is it Needed?

EVMbench is a specialized set of tests based on 120 real high-severity vulnerabilities collected from 40 audits. Most of the data is drawn from open code audit competitions, as well as from the audit of the Tempo blockchain, an L1 network focused on high-performance and low-cost stablecoin payments. The benchmark focuses on three modes of AI agent operation:

• Agents scan smart contract repositories, evaluated based on the completeness of vulnerability detection and potential audit rewards.
• AI modifies vulnerable code, preserving the contract's functionality but eliminating exploitation possibilities. Verification is done through automated tests and attack simulations.
• Agents conduct full attacks to steal funds in an isolated blockchain environment, with verification through transaction reproduction.

The tool uses a Rust-based harness for objective and reproducible evaluation, including deploying contracts in a local Anvil environment and restricting unsafe RPC methods. This allows testing AI in real scenarios close to the blockchain ecosystem.

AI Progress in Cybersecurity

Initial evaluations on OpenAI's frontier models show impressive progress. For example, GPT-5.3-Codex achieved 72.2% success in Exploit mode, a significant improvement over GPT-5's 31.9% just six months ago. However, performance in Detect and Patch modes is lower: AI still struggles with exhaustive audits and preserving functionality when fixing subtle vulnerabilities. These results highlight AI's potential for "defensive" use, auditing and strengthening contracts, but also warn of risks if such tools fall into the wrong hands.

OpenAI and Paradigm Partnership: A Step Toward Secure Crypto

Paradigm, known for investments in projects like Uniswap and Optimism, joined OpenAI to create EVMbench to promote the safe use of AI in crypto. This partnership underscores the growing role of AI in protecting decentralized finance (DeFi). OpenAI also announced future steps: releasing tasks, tools, and the EVMbench framework for open research; expanding the private beta of Aardvark (an agent for security research); free scanning of open-source codebases in partnership with maintainers; and $10 million in API credits through the Cybersecurity Grant Program to accelerate cyber defense, especially for open-source and critical infrastructure.

Community Reactions and Prospects

The EVMbench announcement sparked lively discussions in the crypto community. DeFi developers see it as a tool for raising audit standards, while skeptics fear that improving AI in exploiting vulnerabilities could lead to new hacker attacks. According to open sources, similar benchmarks have already contributed to the growth of the AI-powered security market in blockchain, which is projected to exceed $5 billion by 2030.

This release emphasizes how ChatGPT (as part of the OpenAI ecosystem) and Paradigm have "stood guard" for crypto, combining AI innovations with venture capital expertise.