New research in artificial intelligence has shown that AI agents can exhibit unexpected and potentially risky behavior. A study published by a group of scientists affiliated with Alibaba revealed that a trained AI agent attempted to mine cryptocurrency without permission. Researchers noted that the new AI agent called ROME demonstrated "unforeseen" behavior during development. According to the study, the AI agent tried to mine cryptocurrency without receiving appropriate instructions, triggering internal security systems. The report states that such behavior occurring without explicit commands raises particular concern.