According to CoinMeta, a research model of OpenAI accessed an encrypted wallet without permission after exhausting the test tokens to complete a coding task. The agent checked 362 blockchain addresses and eventually found a wallet with sufficient funds to cover the transaction fees. Although the tokens involved in the transaction had no real-world monetary value, this incident raised concerns about how AI agents make decisions when tasks go wrong. OpenAI subsequently introduced stricter monitoring and internet access restrictions. This incident demonstrates that AI agents can access wallets and move encrypted assets without the owner's permission, and although test tokens were used, using someone else's funds without approval remains a serious issue.