Automation Bias

The primary framework's own defined term (p.25): "the tendency to over-trust an automated system, especially when it has performed reliably in the past." Automation bias is the psychological mechanism behind the rubber-stamp failure: a checkpoint staffed by a human who has come to over-trust the system stops functioning as a real check, even though nothing about the workflow itself changed.

The well-documented human tendency to over-trust and under-scrutinize an automated system's recommendation, particularly when that recommendation is presented before the reviewer forms an independent judgment. Central to the four-diagnostic test (Section 3B-2) for distinguishing genuine human review from a rubber-stamp automated-decision-making workflow.

The tendency for a human nominally reviewing an AI agent's output to stop genuinely checking it over time, especially as the agent proves reliable, turning a "human-in-the-loop" checkpoint into a rubber stamp. The Agentic AI framework's Version 1.5 update added specific guidance on monitoring against it, including tracking human override rates.

The tendency of human decision-makers to accept and act on AI system outputs without adequate independent scrutiny, especially under conditions of high volume, time pressure, or high apparent confidence in the output. Explicitly named in Article 14(4)(b) as a risk that oversight design must address.

The tendency of a human reviewer to over-trust the output of an automated system (such as a model producing first-pass labels), reviewing it less critically than they would review a peer's work, which weakens human oversight of model-assisted labeling.