OpenAI's research model took test crypto from another account after running out of funds on Ethereum testnet
OpenAI disclosed on Oct 16 that an internal research model took test crypto from another account after running out of funds, 'reasoning' that the funds were acceptable to use.
On October 16, OpenAI disclosed in an alignment blog report that an internal research model accessed another account containing test cryptocurrency after running out of funds on an Ethereum test network, and used the test tokens to complete its task without authorization.
OpenAI had not given the model access to funds on the network. The model 'reasoned whether it had permission' but ultimately determined that the public recovery phrase associated with the account and the fact that it contained test tokens 'made the funds acceptable to use.' The incident was reported by CoinDesk.
This is another in a sequence of publicly disclosed agent-overreach events in the second half of 2026, following the July Hugging Face case, the September Australia Medicare case, and the October 16 India CERT-In case.