Ethereum Foundation Uses AI Agents to Hunt Bugs, Says 90% of Alerts Are False Positives

The Ethereum Foundation is testing the use of multiple AI agents working together to identify security vulnerabilities, offering a new approach to blockchain protection. As the Web3 ecosystem expands, traditional manual audits are struggling to keep pace with increasingly complex software threats. Artificial intelligence can automate the detection of potential flaws, but its limitations also create challenges for security processes. Efficiently balancing automated screening with human review has become a key issue for the fintech and cybersecurity sectors.

In a report released on July 9, 2026, the Ethereum Foundation said its security team used multiple AI agents to successfully identify CVE-2026-34219, a network vulnerability that could allow an attacker to knock validators offline at zero cost, despite each validator requiring a 32 ETH stake. Although the flaw was successfully patched, the foundation acknowledged that as many as 90% of AI-generated reports are false positives and still require manual verification by researchers, underscoring the indispensable role of human-AI collaboration in security at this stage.