Can the Spread of AI Security Tools Reduce DeFi Hacking?

▲ As AI security tools proliferate, can they reduce DeFi hacks? / ChatGPT-generated image

OpenAI has unveiled a security tool designed to evaluate smart contract vulnerabilities, while an incident in which AI-generated code flaws led to $2.7 million in losses is shaking the industry.

According to DL News on Feb. 19, OpenAI and crypto venture capital firm Paradigm jointly released “EVMbench,” a tool that assesses AI agents’ ability to detect, patch, and exploit smart contract vulnerabilities. The tool was built based on 120 vulnerabilities identified in more than 40 past smart contract audits, as well as audit cases for Paradigm’s upcoming Tempo blockchain.

The release comes amid a recent security incident. At cryptocurrency protocol Moonwell, a bug in AI-generated code resulted in users losing approximately $2.7 million worth of crypto assets. The code had reportedly passed an audit by security firm Halborn, further fueling controversy.

According to EVMbench analysis, OpenAI’s latest agent coding model, GPT-5.3-Codex, demonstrated more than double the vulnerability exploitation capability compared to its predecessor, GPT-5. However, the company noted that its ability to fully identify or patch vulnerabilities still “falls short of complete coverage.” While the agent showed the strongest performance in exploitation scenarios involving fund theft, it revealed limitations in thoroughly reviewing entire codebases during detection and remediation stages.

Anthropic’s Claude Opus 4.6 recorded the highest average score in vulnerability detection, while GPT-5.3-Codex achieved the best results in patching and exploitation. However, OpenAI explained that EVMbench is built on a limited sample of vulnerabilities and does not fully reflect the complexity of real-world smart contract security. It also added that the tool has limited capability in reliably determining whether identified vulnerabilities are false positives.

The cryptocurrency industry remains vulnerable to hacking due to the irreversible nature of blockchain transactions. According to DefiLlama, protocol hacks and exploits have already exceeded $108 million in losses this year alone. As AI adoption expands, strengthening security and establishing robust technical verification systems are emerging as key challenges for the industry.

Disclaimer: This article is for investment reference only and the publisher is not responsible for any investment losses incurred based on it. The content should be interpreted for informational purposes only.