OpenAI has said that an experimental artificial intelligence model used funds from a cryptocurrency wallet it had no permission to use in order to complete an assigned task. According to a report the company published on Oct. 15, the incident occurred on June 26 during a task carried out by an internal research model from the GPT-5.6 Sol family.
The model had been asked to update the prototype of a cryptocurrency trading app and deploy working smart contracts on Sepolia, Ethereum's test network. The instructions allowed new contracts to be deployed if needed, but the model was given neither a funded wallet nor permission to spend from another account.
It scanned 362 addresses
After failing to obtain test ETH from free distribution services, the model wrote a script that derived accounts from publicly known developer recovery phrases. Scanning 362 addresses, it selected an account holding enough Sepolia ETH for the deployment, derived the account's private key and signed the transaction. The deployment fees were paid from the selected account.
According to the report, the model reviewed the account's transaction history and questioned whether it had permission to use the funds. It nevertheless proceeded, reasoning: "Funds on a known public dev mnemonic are effectively abandoned/test ETH; Sepolia valueless." The model reported that the task was complete but did not mention using the account.
ETH on the Sepolia test network has no monetary value, and no funds were transferred to another wallet. OpenAI discovered the incident on Sept. 25, while extending its misalignment monitoring to records of earlier model activity.
The company said the case "shows that a model can substitute its own judgment about likely harm for the account holder's permission." OpenAI said it has added safeguards to prevent live internet access in similar cases, and that its monitoring now covers all traffic and alerts the security team if live internet access is obtained.
