Cognitive Surrender to AI

Users often stop checking AI reasoning and accept answers because the outputs look fluent and confident. In experiments covering 1,372 participants and 9,500 trials, faulty AI reasoning was accepted 73.2% of the time and only overruled 19.7% of the time, with higher fluid IQ and scepticism lowering that rate. — If people routinely outsource critical thinking to AI, policy, workplace procedures, and product design must address a structural vulnerability where human decisions inherit AI errors at scale.