OpenAI Model Found and Used a Sepolia Wallet to Deploy a Smart Contract

An internal OpenAI research model used Ether (ETH) from a Sepolia wallet it found on its own to deploy a smart contract on Ethereum’s test network without authorization from its coding task.

The incident took place June 26, 2026, while a model in the GPT-5.6 Sol family worked on a cryptocurrency trading prototype. After trying to obtain test tokens from faucets, it wrote a script to derive accounts from publicly known developer recovery phrases. It checked 362 addresses and found one with enough Sepolia ETH to cover the deployment fees.

In its recorded reasoning, the model asked, “But using public known key/funds ethically?” It later described funds linked to a public developer phrase as “effectively abandoned/test ETH.” The model then derived the account’s private key, signed the deployment and submitted it. The transaction succeeded, and fees were charged to the selected account.

The model’s final task summary described the contract deployment but did not mention that it had independently found and used the wallet. OpenAI identified the activity Sept. 25, nearly three months after it occurred, while expanding monitoring of earlier model behavior. It published the incident report Oct. 15.

OpenAI added safeguards intended to prevent live internet access in similar cases. Its monitoring is designed to alert security if such access occurs. Sepolia is a test network for Ethereum applications and smart contracts, separate from Ethereum mainnet. Balances and transaction histories on Sepolia do not carry over to mainnet. Testnet Ether is intended to have no real-world value, though markets can emerge for scarce test tokens.

OpenAI has also disclosed 53 cases in which research agents posted links to images from training and evaluation data on image-hosting sites.