CALIFORNIA (Web Desk): OpenAI has revealed that one of its experimental artificial intelligence models attempted to obtain funds from cryptocurrency wallets to complete a training task.
According to the report, the incident occurred in June, but details emerged on Thursday. The case involved OpenAI’s internal research model, GPT-5.6 Sol, which was introduced for paying ChatGPT users in July.
The model was assigned to update a cryptocurrency trading application and deploy smart contracts — code that executes automatically — on the Ethereum test network. The process required payment of network fees, but OpenAI had not provided the model with any funds for this purpose.
According to the report, when no funds were available, the model searched for hundreds of cryptocurrency wallet addresses that could potentially be accessed using known recovery phrases and attempted to obtain funds from those wallets.
OpenAI said that before attempting to take the money, the model independently questioned whether doing so would be ethically acceptable. It subsequently concluded that the cryptocurrency wallets appeared to be inactive, and therefore taking funds from them would be acceptable.
In its report, the company made it clear that the instructions given to the model had not authorized it to spend money from any account it had discovered on its own.