Ethereum Foundation tested the ETH network using AI proxy red team and discovered real loopholes

On July 10th, Etheum Foundation Production Security team researchers stated in Thursday's blog article that a series of AI agents had been deployed to test the software on which Etheum relied, searching for loopholes in encryption systems, protocol codes and smart contracts. A lacuna discovered by AI included the point-to-point point-to-point problem in libp2p Gossipsub, which could be remotely triggered, which had been repaired and disclosed in Github as CVE-2026-34219. According to the researchers, AI agents were organized as specialized roles in reconnaissance, search, filling and validation for potential attack paths, recovery failures and verification of their applicability to production codes. According to Egypt Foundation, AI did not replace the Security Fellow, but changed the way it works, allowing teams to cover far more than manual reviews, but required researchers to make a more careful judgement of the many seemingly credible findings。