Woofun AI reports that the Ethereum Foundation’s protocol security team employed an AI agent to scan Ethereum network software, uncovering a remotely triggerable crash in the gossipsub messaging system. This flaw forced verification nodes into abnormal computations and automatic shutdowns, requiring manual restarts. The issue is resolved under identifier "CVE-2026-34219".

Researcher Nikos Baxevanis emphasized that the primary challenge was distinguishing genuine vulnerabilities from AI-generated false positives, as the quality of attack descriptions remained consistent regardless of validity. Common false positives included test-build-only crashes, attacks requiring manual dangerous values, and trivial formal verification proofs. While effective for single-point analysis, the AI struggled with multi-step sequential vulnerabilities.