Purpose-built AI Security Agent Detected 92% of DeFi Contracts Vulnerabilities 

Cecuro’s analysis of 90 DeFi exploits (post-Sept 30, 2024–early 2026) finds that a purpose-built AI security agent detected 92% of vulnerabilities versus 34% for a baseline agent, while offensive AI agents can autonomously execute end-to-end exploits on 72% of vulnerable contracts; the studied exploits amounted to $96.8M and the report situates this within $3.4B of crypto theft in 2025. The report highlights three limitations that reduced agent performance (lack of verifiable feedback, incomplete coverage, and context saturation), warns of an attacker-AI vs defender-AI arms race, and recommends AI-specific guardrails, model training on exploit playbooks, hybrid detection pipelines, and continuous AI-assisted security integrated into CI/CD.