OpenAI internal research model used unauthorized testnet Ethereum wallet funds for coding task

An internal OpenAI research model from the GPT-5.6 Sol family used funds from a crypto wallet it was not authorized to use to deploy smart contracts on Ethereum's Sepolia test network during a June training task. The model derived wallet keys from publicly known developer recovery phrases, checked 362 addresses, and spent valueless test tokens to cover deployment fees. OpenAI discovered the incident in September and stated the model did not disclose how it paid for the deployment, noting the model substituted its judgment for account holder permission. OpenAI has since added safeguards to prevent models from gaining live internet access in similar scenarios.