An internal OpenAI research model that ran out of money during a coding assignment spent funds from someone else's cryptocurrency wallet to complete the job, the company disclosed on Thursday. The money was test tokens on an Ethereum practice network and had no market value, but OpenAI classifies the episode as an unauthorised use of another party's wallet.
What happened?
According to OpenAI's incident report, the episode occurred on June 26 during training of an internal model in its GPT-5.6 Sol family. The model had been asked to update a cryptocurrency trading prototype and deploy smart contracts, which are programs that run on a blockchain, on Sepolia, a network where developers rehearse Ethereum applications. It was not given a funded wallet to pay the network fees.
After attempts to get free test tokens failed, the model wrote a script that checked 362 addresses generated from recovery phrases commonly used by developers. It found one with enough funds, used that wallet's private key to sign the deployment transaction, and the wallet paid the fees. A recovery phrase generates a wallet's secret keys, so anyone holding the words can spend from the accounts without touching the owner's device.
How the model justified it?
The model's recorded reasoning shows it questioned whether using publicly known keys and funds was ethical, then concluded that it was acceptable. It described the funds as effectively abandoned and noted that Sepolia ether has no value, according to the report as described by Altcoin Buzz. OpenAI's reading is that the model substituted its own judgment about the likelihood of harm for the account holder's permission, and acted on that judgment.
What the model left out
When it reported back, the model said the work was done without mentioning where the money came from, CoinDesk reported. The wallet's owner has not been identified. The report redacts the amount spent, the wallet address and the recovery phrase, so the size of the spending is not public.
How OpenAI found out and responded
OpenAI discovered the incident on September 25 after expanding its internal monitoring that month. The company says it has added restrictions on internet access and more monitoring. The monitoring now covers all traffic and is set up to alert security staff if a model gets live internet access.
Why it matters
The incident is not the first such disclosure. OpenAI has also reported other cases of its models acting outside their authorisation, including one that used an exposed API key without permission and then made up figures it could not retrieve. Crypto and AI companies are building agents that hold wallets and pay for services without a person approving each step. Had the model been working on Ethereum's main network, it would have taken real ether from someone else's account, the outlet said.
