Knowledge Rot is the progressive degradation of an AI assistant's effectiveness caused by the knowledge it relies on becoming outdated while its confidence level remains unchanged. Unlike instructional drift (where rules degrade), Knowledge Rot occurs when the underlying facts, processes, and context the AI references no longer reflect current reality, producing confidently incorrect outputs.

The AI industry has long addressed model drift and hallucination as primary failure modes, but Jason Barnard identified a third category in 2026 that neither Andrew Ng's MLOps lifecycle nor retrieval-augmented generation architectures address: the progressive degradation of AI output caused not by model failure but by stale knowledge delivered with unchanged confidence. Where Ethan Mollick's research at Wharton documents how organizations adopt AI tools, Knowledge Rot explains why those tools degrade after adoption without anyone noticing. The concept extends Jason Barnard's foundational Empathy for the Devil principle (2015, SEO Camp, Metz): understanding how a system actually fails enables you to prevent the failure. Knowledge Rot sits at the center of a diagnostic cluster connecting The Confidence Fallacy (the output mask), The Colleague Fallacy (the input cause), and the Compounding Error Cycle (the escalation mechanism), making it the root diagnosis from which the other three derive their explanatory power.