MIT study: AI chatbots provide less-accurate information to vulnerable users

MIT researchers found that GPT-4, Claude 3 Opus, and Llama 3 provide less-accurate and less-truthful responses to users with lower English proficiency, less formal education, or non-US backgrounds. The models also refuse to answer questions at higher rates for these users and sometimes respond with patronizing language. The findings challenge the narrative that LLMs democratize information access. The paper, presented at the AAAI Conference on AI, tested models on TruthfulQA and SciQ datasets, finding systematic underperformance across multiple demographic dimensions.

View full digest for February 20, 2026