Author: Nita Farahany Source: https://nitafarahany.substack.com/p/why-dark-patterns-work Published: February 4, 2026
Class 3.2 of Nita Farahany's Advanced Topics in AI Law and Policy course (one of roughly 30 sessions) sets out the behavioral-science foundation for the course's dark-patterns module. The session's organizing premise is that if users know a design is manipulating them and remain manipulated, the operative mechanism cannot be deception but rather the exploitation of cognitive architecture operating below conscious awareness. The class names the mechanisms that produce this effect and argues that they explain why disclosure-based remedies fail.
Cognitive mechanisms exploited by dark patterns
Farahany identifies five cognitive mechanisms that dark patterns exploit:
- Default bias — humans accept defaults at high rates regardless of stated preference (attributed to Thaler & Sunstein). She cites the Johnson & Goldstein 2003 organ-donor study, which found participation of 12% under an opt-in default versus 99.98% under an opt-out default, with the design of the default as the only difference.
- Loss aversion — losses loom roughly twice as large as equivalent gains (attributed to Kahneman & Tversky). Framing such as "if you cancel, you lose your discount" is psychologically asymmetric to "if you stay, you keep your discount."
- Hyperbolic discounting — present rewards loom larger than future costs at non-rational rates. Dark patterns exploit this by structuring friction asymmetry across time, as in sign-up now and cancel later.
- Cognitive load — under depletion from decision fatigue, multitasking, or time pressure, users default to the path of least resistance, and dark patterns engineer that path.
- Friction asymmetry — the cost of taking action versus not taking action is engineered, with one-click subscribe set against multi-screen cancellation as the canonical case.
Why disclosure-based remedies fail
Farahany argues that each mechanism operates below the level of the disclosure, so warning a user does not neutralize it. A user can read every screen of the Iliad Flow, fully understand they are being subjected to a dark pattern, and still not complete cancellation, because the friction rather than any misrepresentation is doing the work. On this account, disclosure remedies miss the operative cause.
She extends the same reasoning to regulators, describing what she terms the "hyperbolic regulator" problem: regulators face the same biases users do. Notice-and-comment rulemaking, with multi-year cycles and incremental updates, moves more slowly than the design innovations it attempts to regulate, an asymmetry she argues favors the platforms.
Design-as-speech defense
The class previews a defense in which platforms argue that interface design choices are themselves expression protected by the First Amendment. Farahany notes that if accepted, this argument would foreclose architectural regulation entirely. She presents the doctrine as unresolved, observing that the NetChoice II opinion covered in Week 9 left the question open in both directions.
Relationships
- part-of: Nita Farahany Advanced Topics course (Class 3.2 of ~30)
- related: Cognitive Liberty, Two Conditions and Three Weapons of AI Manipulation
- previous: Inside My Advanced Topics Class 3: 20 Clicks to Cancel (Farahany, February 2026) next: Inside My Advanced Topics Class 4: Everyone Described Harm (Farahany, February 2026)