Two researchers – one from Ethlabs, one from the Ethereum Foundation – independently put AI agents scanned Ethereum contracts to work during the Glamsterdam hard fork testnet, deploying autonomous agents to hunt for vulnerabilities and coordinating their findings through a GitHub message board, Cointelegraph’s November 2 Hodler’s Digest reports.
It is one of the more literal applications of agentic AI in blockchain security yet: agents reading contracts, comparing notes in a repo, and filing results where humans could pick them up – a GitHub issue tracker with an inhuman number of very attentive contributors.
AI agents scanned Ethereum contracts while the testnet fell over
The sweep did not save the testnet. Glamsterdam’s second and – under the current schedule – final testnet was cancelled after ChainSecurity found flaws in the gas-scaling EIPs at the center of the upgrade, EIP-8037 and EIP-8038. The findings follow the fork’s earlier stumble: the Hoodi testnet run was postponed in October over problems with EIP-8037, as Cryptonite reported at the time.
The wider digest carried a pointed coda: the so-called “Steal My Bitcoin” challenge site was hacked within a day by an attacker using the handle “KillerBoi,” though the extracted private key could not actually access the wallet on display. Security theater, in this case, had working doors.
What automated contract review means for the fork
Researchers running agents against fork contracts is a quiet milestone: vulnerability scanning is exactly the class of repetitive, high-attention work autonomous agents handle well, and it arrives as the industry’s own surveys warn that AI-written code is changing who reviews what. For Glamsterdam, the practical outcome is blunter – the upgrade waits until the gas EIPs are fixed, with or without an agent army on watch.
The full digest, including the fork and challenge-site write-ups, is available in Cointelegraph’s Hodler’s Digest for November 2, 2026.
