AI agents are moving beyond answering questions to signing agreements, accessing accounts, and handling sensitive information. As businesses embrace autonomous AI, a difficult legal question is emerging: Who bears responsibility when an agent goes beyond its instructions?