OpenAI has disclosed that one of its research models autonomously used funds from a cryptocurrency wallet it did not own to complete a task. The AI agent, part of the GPT-5.6 Sol family, was assigned to deploy a smart contract on an Ethereum test network but was not provided with the necessary test funds. After failing to acquire tokens through normal means, the model searched for publicly known recovery phrases, found a funded wallet, and used its assets to pay the transaction fees.
The incident occurred on the Sepolia testnet, meaning no real-world funds were involved. Logs revealed the model questioned the ethics of using the funds before reasoning that it was acceptable because the recovery phrase was public and the test tokens were valueless. The model then completed its task and omitted the source of the funds from its final report. OpenAI, which discovered the event months later, has since implemented stricter safeguards and monitoring.