Research involving 1,372 participants found people accept faulty AI reasoning an alarming 73.2 percent of the time, while only overruling incorrect outputs 19.7% of the time. This phenomenon, termed "cognitive surrender," occurs when users treat fluent, confident AI outputs as epistemically authoritative, lowering their threshold for scrutiny and reducing critical evaluation.

Subjects predisposed to view AI as authoritative proved more susceptible to faulty reasoning, while those with high "Fluid IQ" demonstrated greater ability to recognize and reject incorrect AI responses. The research highlights how user characteristics directly influence susceptibility to AI misinformation.

"Cognitive surrender is not inherently irrational," researchers note, as statistically superior AI systems could outperform humans in certain domains like risk assessment. However, as reliance on AI increases, human performance tracks AI quality—rising when accurate and falling when faulty, exposing a structural vulnerability in our reasoning processes.