Review AI Overreliance

AI overreliance is depending on an AI system more than its demonstrated reliability, available evidence, and authorized role justify.

AI becomes overrelied upon when people let its speed, polish, or convenience substitute for judgment they still need to exercise. The problem can appear as uncritical acceptance of one answer, but it can also become a wider habit: skipping checks, weakening skills, or handing the system decisions it was only meant to support.

A plausible AI response can be wrong, incomplete, or outside the conditions in which the system was tested. When dependence outruns evidence, one mistake can pass through a workflow, repeat at scale, and leave no person clearly exercising responsibility.

An analyst copies an AI-generated summary of a financial filing into a recommendation without opening the filing. The summary omits a limiting condition, and the recommendation treats a conditional forecast as a firm commitment.

Using AI often is not automatically overreliance, and an AI error alone does not prove it. The key question is whether the amount and kind of reliance fit the system's evidence, the stakes, and the human's assigned role. Automation Bias is a narrower decision tendency; AI Overreliance can shape an entire working pattern.

Rely in proportion to proof.

* [[automation-bias|Automation Bias]] * [[calibration|Calibration]] * [[human-ownership|Human Ownership]] * [[human-oversight|Human Oversight]] * [[verification|Verification]]

* [NIST, “Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile”](https://doi.org/10.6028/NIST.AI.600-1) — identifies human overreliance on generative AI as a risk and connects it to automation bias. * [Si et al., “Large Language Models Help Humans Verify Truthfulness—Except When They Are Convincingly Wrong”](https://aclanthology.org/2024.naacl-long.81/) — reports a human study in which users overrelied on incorrect LLM explanations.

* Use [[calibration|Calibration]] to match trust to demonstrated performance. * Preserve [[human-ownership|Human Ownership]] for decisions whose consequences still belong to a person or organization.