Ethereum Foundation used AI agents to red-team test the ETH network and discovered real vulnerabilities.

Golden Finance reported that on July 10, the Ethereum Foundation Protocol Security team researchers said in a Thursday blog post that they have deployed a series of AI agents to test software relied upon by Ethereum, searching for vulnerabilities in encryption systems, protocol code, and smart contracts. The vulnerabilities discovered by the AI agents include a remotely triggerable panic issue in the libp2p gossipsub peer-to-peer layer used by Ethereum consensus clients, which has been fixed and disclosed on GitHub as CVE-2026-34219.

The researchers stated that the AI agents are organized into specialized roles such as reconnaissance, hunting, patching, and verification, used to find potential attack paths, reproduce failures, and verify whether they apply to production code. The Ethereum Foundation said that AI has not replaced security researchers but has changed the way they work, enabling the team to cover far more scope than manual review, while requiring researchers to more carefully judge a large number of seemingly plausible conclusions.