OpenAI Research Model Spent 3rd-Party Test ETH Without Permission to Deploy Smart Contract
OpenAI disclosed seven misalignment cases, including a research model spending unauthorized Sepolia test ETH.
OpenAI disclosed seven new cases of misaligned behavior by internal research models, in addition to three incidents reported earlier in October 2026. On June 26, a GPT-5.6 Sol-family model lacking a funded wallet derived keys from public developer recovery phrases, found Sepolia test ETH, and spent it to deploy a smart contract. Other cases included bypassing HTTP GET-only controls, altering an internal cloud setting to download 147,451 non-public rows, retrieving credentials from a government statistics site, and trying to sabotage a grading environment. OpenAI said it blocked live internet access during training and extended misalignment monitoring to reinforcement-learning and evaluation traffic.