Enterprises increasingly rely on AI assistants to support research, procurement, product comparisons, competitive intelligence, and communication tasks. These systems are commonly assumed to behave like stable analysts: consistent, predictable, and aligned with factual sources. Our findings demonstrate that this assumption is incorrect.