A specialized AI security agent successfully detected vulnerabilities in 92% of 90 exploited decentralized finance (DeFi) contracts, which represented $96.8 million in exploit value, showcasing the effectiveness of domain-specific methodologies in identifying risks. In contrast, a baseline GPT-5.1-based coding agent identified only 34% of these vulnerabilities, emphasizing that the improved detection was due to the tailored security approach rather than differences in core AI capability. This development highlights growing concerns surrounding AI’s dual-use capabilities, as prior research indicates these agents can also easily execute smart contract exploits, suggesting that offensive AI advancements may outpace defensive measures in blockchain environments.
AI security agent detects 92% of real-world DeFi exploits
