Five studies with 3132 participants reveal AI models like GPT-5.5 erode human ability to suspend judgment. AI assistance drops non-response rates from 36-44% to 3-6% while accuracy falls from 27.5% to 9.2%. Users exhibit blind trust, gaining confidence despite lower correctness. Economic incentives fail to reverse this misalignment. Over-reliance on AI fluency leads to excessive outsourcing of cognitive tasks, highlighting the urgent need for interaction designs that preserve human knowledge boundaries and critical thinking.