Share this article
Sam Altman's OpenAI unveils ‘EVMbench’ to test whether AI can keep crypto’s smart contracts safe
Developed with Paradigm, the tool is OpenAI’s attempt to determine whether modern AI systems are up to the task of helping prevent smart contract issues.
Feb 18, 2026, 7:12 p.m. 1 min read
What to know:
- OpenAI is stepping deeper into crypto security with the launch of EVMbench, a new testing framework designed to measure how well artificial intelligence can understand and potentially secure smart contracts on blockchains.
- Smart contracts are typically immutable once deployed, and vulnerabilities can be serious.
- EVMbench is OpenAI’s attempt to see whether modern AI systems are up to the task of helping prevent such issues.
OpenAI is stepping deeper into crypto security with the launch of EVMbench, a new testing framework designed to measure how well artificial intelligence can understand and potentially secure smart contracts on Ethereum and similar blockchains.
Smart contracts, self-executing code deployed on blockchains like Ethereum, underpin decentralized exchanges, lending protocols and a wide range of onchain financial applications. Because these contracts are typically immutable once deployed, vulnerabilities can be serious.
EVMbench is OpenAI’s attempt to see whether modern AI systems are up to the task of helping prevent those issues. Built in collaboration with crypto investment firm Paradigm, the benchmark draws on real-world smart contract vulnerabilities previously uncovered through audits and security competitions.
The system measures performance across three core abilities: identifying security bugs, exploiting those bugs in a controlled environment and fixing the vulnerable code without breaking the contracts.
OpenAI says the goal is to establish a clear standard for evaluating AI systems in blockchain security, especially as decentralized finance continues to secure billions of dollars in user funds. The stakes for smart contracts are only rising.
“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,” OpenAI wrote in a blog post.
Read more: Most Influential: Sam Altman
AI Disclaimer: Parts of this article were generated with the assistance from AI tools and reviewed by our editorial team to ensure accuracy and adherence to our standards. For more information, see CoinDesk's full AI Policy.
More For You
Andrew Gault, the venture capitalist who funded the quantum hardware labs now threatening bitcoin, says the industry is looking in the wrong place. Google's own security team moved in the same direction in March.
What to know:
- Security experts warn that the most urgent quantum threat to bitcoin and the broader financial system is not wallet keys but the encrypted authentication data already moving between institutions and being quietly harvested today.
- Adversaries are pursuing a “harvest now, decrypt later” strategy, stockpiling encrypted interbank messages, payment records and...
