An AI chatbot was consulted about military action, raising questions about security risks, sycophancy and hallucination.