AI Accidents: A Rogues’ Gallery
Inspired by the nineteeth-century police “Rogues’ Galleries” that helped the public recognize criminals, this collection documents the strange, funny, and revealing ways generative AI can mislead us.
Stories of machine mischief, human confusion, and unexpected delight.
Each case captures a moment where fluent language collides with human judgement. Some are obvious mistakes. Others are surprisingly persuasive.
Together they form a training set for learning how AI behaves and how we respond to it.
Share your own encounter and help expand the gallery.
⚠️ These examples capture generative AI behaviour at a moment in time. As systems evolve. some details may become outdated. The patterns they reveal, however, remain valuable for understanding human judgement in the age of AI.
#003: The Sneak
My take on GenAI’s version of Dr. Harry Brignull’s “Sneaking” deceptive pattern, the Sneak happens when GenAI quietly introduces a new idea, assumption, or angle that wasn’t in your prompt—but now feels like it was.
#002: The Mirror Trap
The Mirror Trap happens when GenAI reflects your assumptions, language, or implied beliefs right back at you—without question. It sounds affirming and collaborative, but it’s not evaluating your idea—it’s amplifying it.
#001: The Measured Hallucination
The Measured Hallucination is when GenAI invents facts, names, or events that sound real—and presents them in a thoughtful, considered way. It can be even more deceptive than an obvious error.