OpenAI, crypto investor Paradigm launch AI tool for smart contracts

OpenAI and Paradigm have announced the launch of EVMbench, a benchmark designed to evaluate AI agents’ ability to detect, patch, and exploit smart contract vulnerabilities.

The tool comes amid recent incidents where DeFi protocols, including Moonwell and CrossCurve, experienced smart contract exploits, with CrossCurve losing around US$3 million across multiple networks.

EVMbench tests AI capabilities in various modes, such as vulnerability detection and contract modification, using a dataset of 120 vulnerabilities from 40 audits, including open-code competitions and Tempo blockchain security reviews.

The initiative aims to assess AI systems’ potential to improve smart contract security in decentralized finance.

🔗 Source: The Block

🧠 Food for thought

Implications, context, and why it matters.

Existing security tools cover less than half of known vulnerabilities

EVMbench goes beyond reacting to recent DeFi smart contract exploits. It targets a wider shortfall in automated smart contract security.
A systematic review counted 192 unique smart contract vulnerabilities. It found 113 missing from coverage across 219 detection tools 1.
Many scanners cluster around a few familiar issues. For instance, 130 products say they catch “reentrancy” attacks (a class of bug where a contract can be tricked into calling back into itself in an unsafe way), yet only 16 vulnerabilities are covered by ten or more tools 1.
EVMbench tests whether an AI agent can handle harder cases, including interactions that span multiple smart contracts 2.

A new benchmark for AI could reshape software development and insurance

A standardized yardstick could give AI-powered code analysis tools (software that reviews code to find security flaws) a consistent “quality score.” That could push wider use, similar to how automated test suites fit into software builds today 2.
The benchmark also checks whether an AI can explain flaws in human-readable formats. That may shorten debugging cycles for smart contract developers and security auditors 2.
More dependable AI security assessments could shape DeFi insurance by improving risk-pricing models for smart-contract-driven assets and protocols insurers underwrite 2.
The modular setup may also act as a blueprint for other fields that need rigorous testing of AI agents on specialized, high-stakes code 2.