It has been reported that when users ask questions about "Minecraft" to the AI search feature integrated into the official U.S. government information website "America.gov," it generates extremely bizarre responses that deviate significantly from the context. Experts point out that this phenomenon is not a temporary system glitch, but rather a behavior stemming from the characteristics of the AI model and its training data.
America.gov has implemented an interactive AI-powered search function to allow citizens to access government information quickly. In the recently reported case, the AI generated outputs lacking logical consistency for specific topics. This is believed to be caused by hallucinations unique to Large Language Models (LLMs) and model biases in prompt processing.
This incident highlights the difficulties of implementing AI in domains that require high reliability and accuracy, such as government platforms. Strengthening guardrails to ensure optimal and accurate answers to a wide variety of questions, along with establishing monitoring systems for model outputs, will become critical focal points for the public sector's adoption of AI moving forward.
At this time, no official patch or announcement has been made by the platform operators. As the rapid societal implementation of AI progresses, governance frameworks—specifically how to guarantee and continuously control the validity of AI responses—will face even greater scrutiny.