The Ethereum Foundation has revealed that its security team successfully identified a remotely exploitable vulnerability in Ethereum validator node software using AI agents. This flaw, found in the gossipsub protocol, could cause nodes to crash and go offline until manually restarted. The vulnerability has been patched and registered as CVE-2026-34219. Despite AI's ability to detect such vulnerabilities, researchers emphasize the necessity of human oversight to distinguish real threats from false positives. AI-generated reports often include detailed narratives and attack simulations, but human review is crucial to verify their accuracy. The Foundation noted common false positives include crashes only in test environments, attacks requiring manual data modification, and mathematically valid but practically irrelevant conclusions. The Foundation plans to continue leveraging AI for risk detection but acknowledges the need for human intervention in complex attack scenarios, such as those seen in recent DeFi exploits. These often involve intricate combinations of normal actions that AI struggles to identify as malicious.