Imagine having a personal AI travel agent that crafts the perfect itinerary based on your interests, budget, and identity. Large language models (LLMs) are making this dream a reality, but are they truly ready for prime time? New research explores the capabilities and biases of open-source LLMs like Gemma-2 and Llama-3 when tasked with travel planning. Researchers found that while these AI agents can create seemingly personalized trips, they often rely on stereotypes, associating certain ethnic groups with specific cuisines and landmarks. For example, African Americans were frequently recommended soul food and historically Black neighborhoods, while Asians received suggestions for noodles, sushi, and Chinatowns. While seemingly innocuous, these biases highlight how LLMs can reinforce cultural assumptions. Interestingly, LLMs also appeared sensitive to safety concerns, suggesting welcoming and inclusive destinations for gender minority groups and emphasizing female-only accommodations for solo female travelers. By using a clever “fairness probing” technique with machine learning classifiers, the study reveals how these LLMs reflect and potentially amplify existing societal biases. While the initial results raised concerns, further analysis using “stop words” suggested that these biases might not be as deep-rooted as initially feared. However, the research also uncovered unexpected “hallucinations” where the AI fabricated non-existent restaurants or inserted irrelevant dates, reminding us that these powerful tools are still prone to errors. The study's findings are a crucial reminder that while AI travel agents hold immense promise, we need to carefully address their biases and ensure they offer truly personalized and unbiased travel experiences for everyone. Future research will explore other LLMs and travel-related tasks, like handling complaints, to better understand how we can harness the power of AI for a more inclusive and enjoyable travel future.