AI Detects 92% of Real-World DeFi Vulnerabilities, Study Finds

A recent study reveals that specialized AI systems have successfully identified vulnerabilities in 90 DeFi smart contracts with a detection accuracy of 92%. According to Bitalk News, this is a significant improvement compared to GPT-5.1-based systems, which only detected 34% of vulnerabilities, involving a total of $7.5 million. The specialized AI utilized domain-specific security methods, surpassing the limitations of traditional AI models. The study highlighted that the total losses associated with the smart contracts amounted to $228 million, with the vulnerabilities detected by the specialized system valued at $96.8 million. These findings have raised concerns about the challenges AI poses in accelerating crypto-related crimes. Various studies indicate that AI can exploit smart contract vulnerabilities at a low cost, rapidly enhancing attack capabilities. The research data and evaluation framework have been made available on GitHub as open-source, although the complete tools have not been released to prevent misuse.