Specialized AI detects 92% of real-world DeFi exploits

New research claims specialized AI dramatically outperforms general-purpose models at detecting exploited DeFi vulnerabilities. What to know : A purpose-built AI security agent detected vulnerabilities in 92% of 90 exploited DeFi contracts ($96.8 million in exploit value), compared with 34% and $7.5 million for a baseline GPT-5.1-based coding agent running on the same underlying model. The gap came from domain-specific security methodology layered on top of the model, not differences in core AI capability, according to the report. The findings come as prior research from Anthropic and OpenAI shows AI agents can execute end-to-end smart contract exploits at low cost, accelerating concerns that offensive AI capabilities are scaling faster than defensive adoption.