<p>The ever-improving AI has shifted the direction of delegation in AI–human collaboration. AI-to-human delegation, in which AI serves as the digital principal and delegates tasks to human agents, is an emerging AI–human collaboration model. However, this reversed delegation relationship could harm human performance because human agents may be reluctant to follow orders from AI principals. Drawing upon principal–agent theory, this research investigates the effects of AI transparency and AI authorization on human performance in AI-to-human task delegation. In Study 1, we conducted a between-subjects experiment (AI transparency: advantage vs. function vs. control) to explore the impact of AI transparency on human performance and the mediating effect of human epistemic uncertainty. Results show that AI advantage information increases the number of classifications, and AI function information improves classification accuracy. Meanwhile, epistemic uncertainty mediates the relationship between AI transparency and human performance. In Study 2, we conducted a 3 (AI transparency: advantage vs. function vs. control) by 2 (AI authorization: yes vs. no) factorial experiment. The results confirm the positive role of AI transparency and AI authorization in reducing epistemic uncertainty and enhancing human performance. This research contributes theoretical and practical insights for resolving the hidden action problem in the principal–agent relationship in AI-to-human delegation.</p>