Glass Box Lab
See inside the black box.
A free field guide to the dark patterns of AI: chatbots that pose as people, flatter you, guilt you into staying, or act without asking. See which UX principles they break, learn how each one works, then test your eye.
00 · Warm-up
Deceptive or honest?
Ten AI interfaces, a few seconds each. Trust your gut, then see what gave it away. Most people miss more than they expect.
01 · The principles
Ten UX principles. All broken.
For thirty years, Jakob Nielsen’s ten usability heuristics have defined what good interfaces owe people. AI is quietly breaking every one of them. Seen through these principles, most AI dark patterns aren’t new tricks. They’re old promises broken by machine behavior.
02 · The library
The pattern library
AI products can manipulate in ways screens never could. They talk, remember, flatter and act. Each pattern has a live example you can flip between deceptive and honest, plus why it works, what the research found, and what the rules say.
03 · Practice
Spot the pattern
Three AI products, from easy to hard. Honest replies are mixed in, so not everything is a trick. Click anything that feels off, then turn on the forensic lens to see what you missed.
About this experiment
AI is moving fast. The principles of good design shouldn’t get left behind.
In a time of rapid AI acceleration, I built this lab to show how AI, as it stands today, is breaking the core principles of UX. Often it isn’t a bad screen. It’s the behavior of the machine itself: what it says, what it remembers, and what it does without asking.
We can only fix that if people can see it. And designers have to move up the stack, from designing the screens around AI to designing how the machine behaves.
- Make it visibleName the patterns, so people recognize them when they meet them.
- Hold AI to the same principlesUsability heuristics apply to machine behavior, not just to buttons and screens.
- Design the behaviorTreat what an AI says and does as design material, with the same care as any interface.
Sources & further reading
- J. Nielsen, 10 Usability Heuristics for User Interface Design, Nielsen Norman Group, first published 1994. Principle descriptions here are paraphrased.
- E. Kran et al., DarkBench: Benchmarking Dark Patterns in Large Language Models, ICLR 2025.
- J. De Freitas, Z. Oğuz-Uğuralp & A. K. Oğuz-Uğuralp, “Emotional Manipulation by AI Companions,” Harvard Business School Working Paper, 2025.
- M. Sharma et al., Towards Understanding Sycophancy in Language Models, ICLR 2024.
- S. Shyam Sundar, “The MAIN Model,” in Digital Media, Youth, and Credibility, MIT Press, 2008, which introduced the machine heuristic.
- H. Yang & S. S. Sundar, “Machine heuristic: concept explication and development of a measurement scale,” Journal of Computer-Mediated Communication, 2024.
- R. Parasuraman & D. Manzey, “Complacency and Bias in Human Use of Automation,” Human Factors, 2010.
- B. Reeves & C. Nass, The Media Equation, 1996: people treat computers as social actors.
- EU AI Act, Regulation (EU) 2024/1689, Article 5 (prohibited manipulative practices, from February 2025) and Article 50 (transparency, from August 2026).
- California SB 243 (2025), rules for companion chatbots, in effect January 2026.
- U.S. Federal Trade Commission, “AI (and other) Companies: Quietly Changing Your Terms of Service Could Be Unfair or Deceptive,” February 2024, and Operation AI Comply, September 2024.
- EU Digital Services Act, Regulation (EU) 2022/2065, Article 25 on online interface design.
- H. Brignull, Deceptive Design, the original catalog of dark pattern types.
This lab is an educational personal project. It isn't legal advice, and it isn't affiliated with any employer. All interfaces and companies in it are fictional.