Slide 4 · Definition Part 2
Overreliance: the user’s half.
Why users trust output they should verify.
The Human Half of the Problem
Hallucination explains why the model lies. Overreliance explains why the lie causes harm. Both are required for a bad outcome.
🏛Authority bias
AI systems are perceived as knowledgeable, neutral, and more reliable than a search result. Users extend trust they would not give a random stranger on the internet.
✍Fluency signals credibility
The model writes in polished, professional language. Humans associate fluent, well-structured text with accuracy. The Mata attorney later said the fabricated cases “seemed real” — because they were written exactly like real cases.
⏱Productivity pressure suppresses verification
If using the LLM saves two hours, pausing to verify each output feels like giving up the benefit. Deadline pressure reinforces this: skip the check, ship faster.
🔔No uncertainty signal in the output
Unlike a search result (where zero results found is obvious), an LLM always produces something. There is no visual indicator that says “I am guessing here.”
The Compounding Effect
Overreliance turns a quality issue into a security risk. A wrong answer that nobody trusts causes no harm. A wrong answer that goes directly into a court filing, a medical decision, or a codebase is LLM09.