Definition
Plain language
An AI's tendency to stop checking the moment it sees a clean, direct-looking answer.
As stated in the literature
A failure mode where an agent's stopping policy treats a fluent, answer-shaped document as permission to terminate the investigation, ending evidence-gathering before reconciling it against retrieved ground truth.
Why it matters: It causes AI systems to accept plausible-looking claims without cross-checking, letting confident falsehoods slip through as final answers.
For example, an AI finds a page that neatly states an answer and immediately stops looking, never noticing its other sources point elsewhere.
Heard on the show
“The authors call it verification inertia.”Episode 224 — The AI Agent That Found the Truth and Typed the Lie Anyway