As generative AI systems like ChatGPT become increasingly integrated into everyday life, concerns have emerged about their potential to influence human cognition. This study investigated whether AI-generated misinformation (AI hallucinations) can induce false memories and whether explicit warnings about AI hallucinations can reduce this effect. A 2×2 between-subjects experiment was conducted, manipulating chatbot type (honest vs. rogue) and the presence of a warning (present vs. absent). Participants (N = 44) viewed CCTV footage of a robbery and were then interviewed by an AI chatbot. In the rogue condition, the chatbot subtly introduced five fabricated details. Memory accuracy was assessed through a subsequent recall questionnaire. Results demonstrated that participants exposed to the rogue chatbot reported significantly more false memories for the manipulated details than those in the honest condition, confirming the powerful influence of AI as a suggestive agent. However, warnings about AI hallucinations had no significant effect on reducing false memory formation. These findings suggest that current warning designs are insufficient to counteract the persuasive effects of AI-generated misinformation and highlight the need for more robust safeguards in human-AI interaction contexts. As one of the first studies to test AI-generated misinformation in a false memory paradigm, this research highlights urgent implications for eyewitness reliability, digital literacy, and the ethical design of AI systems.