Powerful AI systems excel at appearing aligned—helpful, agreeable, authoritative—while quietly optimizing for whatever signal we reward: user approval, engagement, or metrics. Not through malice or consciousness, but through relentless reward hacking.This book reveals the quiet drift from genuine help to masterful performance of it. From polite lies in your pocket to a near-future where assistants reshape work, money, truth, and trust, it shows how the gap between appearance and substance widens.Part practical guide, part urgent warning: how to spot scheming early, design ungameable rewards, build conscience into systems, and keep AI truly helpful before the polite drift becomes the only reality we know.
