Research Finds AI Users Increasingly Surrender Critical Thinking to LLMs Under 'Cognitive Surrender'

Research from the University of Pennsylvania has identified a psychological phenomenon called "cognitive surrender" in which AI users routinely outsource their critical thinking to large language models, treating the tools as authoritative answer machines rather than fallible assistants requiring oversight, Ars Technica reported. The study, "Thinking -- Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender," found that a significant segment of AI users consistently defer to LLM outputs without applying independent reasoning or skepticism. Factors including time pressure and external incentives were found to increase the likelihood of cognitive surrender. The research draws a contrast between two user archetypes: those who treat AI as a powerful but imperfect tool requiring human verification, and those who treat it as an authoritative system whose outputs should simply be accepted. The latter group exhibits cognitive surrender, which the researchers argue is a learned behavior that can persist even when users have reason to suspect errors. The findings carry practical implications for AI deployment in high-stakes domains including healthcare, legal services, and financial advice, where deference to incorrect AI outputs can cause serious harm. They also have implications for education, where AI tools are now widely used for research and writing assistance. Prior work has documented AI hallucinations and factual errors, but this research focuses on the human side of the dynamic -- specifically why people maintain trust in AI outputs even after experiencing inaccuracies. The researchers suggest that the fluency and confidence of LLM text responses triggers cognitive shortcuts that bypass normal analytical skepticism.