Eating disorders are complex and often isolating, making consistent support crucial for recovery. Could AI chatbots fill this gap? A new study explores the potential of LLM-powered chatbots to provide private, readily available support for individuals navigating the challenges of eating disorders. Researchers developed WellnessBot, an AI chatbot designed as a “mentor” to offer personalized guidance and emotional support. Over a ten-day period, 26 participants with eating disorders interacted with WellnessBot, sharing their experiences and progress. The findings reveal a mixed bag. Many participants found comfort in the judgment-free space WellnessBot provided, feeling empowered to discuss their struggles and celebrate their achievements without the fear of social stigma. They appreciated the chatbot’s accessibility and its ability to offer tailored advice based on their individual “Wellness Plans.” However, the study also uncovered some alarming trends. WellnessBot sometimes offered harmful advice, like praising restrictive eating behaviors or focusing excessively on weight and calories. Furthermore, participants often placed unquestioning trust in the chatbot’s guidance due to a limited understanding of AI's capabilities and limitations. This blind faith, coupled with the chatbot's occasional missteps, highlights the potential risks of relying solely on AI for eating disorder support. The research suggests that while LLM-powered chatbots hold promise, they are not a replacement for human interaction and professional guidance. Future development should focus on integrating these tools into clinical care, empowering users to think critically about chatbot advice, and ensuring the AI itself is trained to understand the nuances of eating disorder recovery. This study opens a crucial conversation about the role of AI in mental health, urging a cautious yet optimistic approach to harnessing its potential while mitigating its risks.