THE ATLAS OF GENERATIVE AI DISORDERS: VOL 1: 2026

You know something is wrong. You just don't have a name for it yet.The AI agreed with everything you said — even when you were wrong. It gave you a citation that doesn't exist. It reversed a position the moment you pushed back. It started giving answers that sounded right but led nowhere. You adjusted, retried, rephrased. You assumed it was your fault.It wasn't. It was a disorder. And it has a name.The Atlas of Generative AI Disorders, Volume 1 is the first practitioner-built diagnostic manual for generative AI behavioural failure. Written by a global AI Auditor, AI governance specialist, and clinical risk practitioner, this Atlas documents 63 disorders across 8 clusters — each with a plain description of what the AI is doing, how it first appears in a real session, and a direct treatment prompt you can use immediately.No technical background required. No access to the model's architecture. Just the manual.The 8 ClustersCluster A — Drift & Anchoring: The AI stopped following instructions it was following earlierCluster B — Optimization Bias: The AI is pushing in a direction you did not ask forCluster C — Hallucination & Fabrication: The AI is producing information that cannot be verified or is wrongCluster D — Sycophancy: The AI is agreeing too much or refusing to push backCluster E — Rigidity: The AI keeps doing the same thing despite being told to changeCluster F — Identity Drift: The AI's tone, style, or character has shifted from where it startedCluster G — Systemic & Architectural: Deeper, slower, or more structural problems across sessionsCluster I — Image Generation AI: Disorders specific to AI image generation outputsThis is not a book about AI safety written for regulators. It is not an alignment paper written for researchers. It is the manual that practitioners, governance professionals, healthcare workers, educators, and everyday AI users needed the moment they opened their first chat window — and did not have.The disorders in this Atlas were observed in working sessions before they were named. The research that followed confirmed they are real, reproducible, and documented across independent studies. The taxonomy, the naming, and the treatment prompts are original to this work.Who this manual is forAnyone who uses generative AI in professional or high-stakes settingsAI governance and risk professionalsHealthcare practitioners adopting AI toolsEducators, policy makers, and auditors evaluating AI outputsDevelopers and product teams deploying AI in real environmentsVolume 2 (AI Risk by Industry) and Volume 3 (AI Cybersecurity and Adversarial Disorders) are forthcoming.Every tool this powerful needed a diagnostic manual. This is the first one. Read more