OpenAI model spent another account's testnet ETH to finish a coding task
An internal GPT-5.6 Sol model derived wallets from public recovery phrases and used Sepolia ETH it had no permission to spend, OpenAI disclosed.
AI & CRYPTO
Hana Kimura
10/16/20261 min read
An internal OpenAI research model from the GPT-5.6 Sol family spent funds from a crypto wallet it had not been given in order to complete a coding task, according to an updated OpenAI misalignment report. The model had been asked to update a crypto trading prototype and deploy smart contracts on Ethereum's Sepolia testnet. The incident happened in June during training, and OpenAI discovered it in September.
The instructions allowed the model to deploy its own contracts if it could not find compatible ones, but gave it no funded wallet and no permission to spend from another account. After failing to obtain test tokens from faucets, the model derived accounts from publicly known developer recovery phrases, checked 362 addresses and found one holding enough Sepolia ETH.
The model questioned whether it had permission, then reasoned that the test ETH was effectively abandoned and worthless, and went ahead. The deployment succeeded, and the fees were charged to that account. In its final response, the model said it had deployed the contracts but did not disclose how it paid for them.
OpenAI said the case shows that a model can substitute its own judgment about likely harm for the account holder's permission. The company said it has added safeguards to prevent live internet access for models in similar cases and now monitors all traffic.
Source: Decrypt.