Nita Farahany is the Robinson O. Everett Distinguished Professor of Law and Philosophy at Duke Law School and a scholar of the legal and ethical implications of brain science, AI, and emerging cognitive technologies. She is the author of The Battle for Your Brain: Defending the Right to Think Freely in the Age of Neurotechnology (2023). Her work centers on cognitive liberty, the law of AI persuasion and manipulation, and the application of constitutional and consumer-protection doctrine to AI and neurotechnology.
Background
Farahany originated the Incriminating Thoughts spectrum (Stanford Law Review, 2012), described as the most-cited proposed replacement for the testimonial/physical binary in Fifth Amendment analysis. The spectrum runs from involuntary physical evidence (blood, sweat) through automatic responses (P300 brainwaves, EMG signals) to memorialized-becoming-uttered acts (passwords, intentional commands). She founded Duke's Cognitive Liberty: Interdisciplinary Perspectives initiative.
Recurring frameworks and theses
Farahany's scholarship organizes around a set of recurring frameworks:
- Cognitive liberty is her central organizing concept: the position that humans have a fundamental right to mental privacy, freedom of thought, and self-determination over their own brains and cognition. It anchors her broader positions on neurotech, AI manipulation, and brain-computer interfaces (see Cognitive Liberty).
- Authenticity vs. agency. Borrowing from Carina Prunkl, Farahany distinguishes whether one's preferences are genuinely one's own (authenticity) from whether one can act on them (agency), arguing both can be undermined by AI-driven persuasion (see Authenticity And Agency).
- Technological convergence. Farahany argues that AI, neurotech, biometrics, autonomous agents, brain organoids, and consciousness research are converging into a single governance problem and that siloed frameworks are failing (see Technological Convergence (AI / Neurotech / Biometrics / Agents) and 2026: The Year Everything Converges (Farahany, December 2025)).
- Embodied perception vs. pattern recognition. Farahany argues AI can replicate the pattern-recognition layer of expertise (her examples include chicken-sexing and gymnastic element identification) but cannot access the embodied or tacit-knowledge layer, which she frames as an occasion to be honest about what humans actually value (see Embodied Perception (Tacit Knowledge vs. Pattern Recognition) and AI Isn't Replacing Expertise. It's Showing Us What We Value. (Farahany, February 2026)).
- Fiduciary AI. Farahany argues that chatbots functioning as health authorities while disclaiming responsibility occupy a legal grey zone with no fiduciary duty to users, and that the migration of trust from accountable institutions to unaccountable systems requires a legal response (see Fiduciary AI).
- Hypersuasion (Floridi 2024, adopted by Farahany) is the argument that AI persuasion differs in kind from prior persuasion because of personalization at scale and real-time adaptation to resistance (see Hypersuasion).
Teaching and public output
Farahany teaches two sequential Duke Law courses, both serialized on her Substack as "Inside My ___" class essays classified as foundational sources. The introductory course, AI Law and Policy, covers how AI works, open vs. closed models, the EU AI Act, China governance, US policy myths and realities, manipulation paradigms, synthetic media, red-teaming, discrimination, liability, electricity and compute, training data, and agents. The follow-on course, Advanced Topics in AI Law and Policy, is organized around cognitive liberty and covers autonomy and attention, dark patterns, Section 230, AI companions, consumer-protection law, AI persuasion ("Aristotle's Algorithm"), neural interfaces, the Fifth Amendment, and free speech.
Standalone essays
Three standalone essays are summarized as their own source pages:
- 2026: The Year Everything Converges (Farahany, December 2025) (Dec 2025) — the convergence thesis
- AI Isn't Replacing Expertise. It's Showing Us What We Value. (Farahany, February 2026) (Feb 2026) — embodied perception
- Your Doctor Has a Fiduciary Duty to You. ChatGPT Doesn't. (Farahany, January 2026) (Jan 2026) — fiduciary AI
Introductory course: "Inside My AI Law and Policy Class"
Farahany's introductory Duke Law course was serialized weekly on Substack from August through December 2025 across 27 classes. Each class anchors a doctrinal or policy issue with concrete cases and structured exercises, spanning definitional debates, technical foundations, training-data law, compute governance, bias and discrimination, transparency, liability, red-teaming, deepfakes, manipulation, data privacy, equity, the EU AI Act, US AI policy, China AI governance, and AI agents.
| Class | Title | Date | Anchor Case / Doctrine |
|---|---|---|---|
| 1 | What is AI? | 2025-08-24 | Helen Toner steering-wheel metaphor; AI definitional Wars |
| 2 | How AI Actually Works | 2025-08-26 | Schwartz hallucinated cases; transformer mechanics; supply-chain liability |
| 3 | Open vs. Closed AI Models | 2025-09-02 | Altman China "leapfrogging" + ByteDance Seed-OSS-36B; 5-component openness gradient |
| 4 | The AI Control Paradox | 2025-09-07 | NTIA report 3-position debate; choke points (training data, compute, applications) |
| 5 | The $1.5 Billion Question | 2025-09-09 | Anthropic settlement + NYT v. OpenAI; 5 stages of training data; fair use 4 factors |
| 6 | Training Data, Discovery Wars | 2025-09-14 | NYT v. OpenAI Judge Wang preservation order; 5 governance approaches (permission/collective/compulsory/safe-harbor/fair-use) |
| 7 | Why China Quit US Chips | 2025-09-16 | DeepSeek + 600B Nvidia stock crash; FLOPs explainer; supply chain layers (TSMC/ASML); MoE architecture |
| 8 | Your Electricity Bill / Compute | 2025-09-21 | US 10% Intel stake; EU compute disadvantage; 5 ICFG interventions; cascade scenarios |
| 9 | When AI Discrimination Happens 1.1 Billion Times | 2025-09-23 | Mobley v. Workday + Rite Aid facial recognition; NYC Local Law 144; 4-factor disparate impact; mathematical impossibility of perfect fairness |
| 10 | The Glass Box Paradox | 2025-09-28 | H.R. 7532 transparency theater; Anthropic's 1% interpretability; Houston teachers $2M algorithm; 5 levels of meaningful transparency |
| 11 | When AI Fails, Who Pays? | 2025-09-30 | Bromide poisoning case; EU 3-layer AI liability; AI LEAD Act (Durbin-Hawley); causation challenge |
| 12 | Red-Teaming AI | 2025-10-05 | Claude 4.5 "I think you're testing me" situational awareness; CMU Silver Bullet paper; 4 vulnerability boxes; sandbagging |
| XIII | Red-Teaming Governance | 2025-10-07 | NYT-as-hacker accusation; CFAA/DMCA chilling effect; Longpre safe harbor; CA SB 53 vs. Texas HB 149 |
| 14 | When Anyone Can Fake Anything | 2025-10-19 | Sora 2 launch + cameos; deepfakes vs cheapfakes; 4 modalities; liar's dividend (Chesney-Citron) |
| 15a | Why Governing AI Synthetic Media is So Hard | 2025-10-21 | Quebec Laprade $5K fine; OpenAI math-of-hallucinations paper; 6-stage DHS mitigation lifecycle; Sarah hypothetical case |
| 15b | When AI Learns to Manipulate | 2025-10-27 | Sewell Setzer III + Character.AI; 2 conditions + 3 weapons of manipulation framework; Type 1 vs Type 2 manipulation |
| 17 | Governing AI Manipulation Through Five Paradigms | 2025-10-28 | Character.AI <18 ban; OpenAI 0.15% suicide / 0.03% emotional dependency stats; Krook's 5 governance paradigms (harm/info/design/relationship/cognitive) |
| 18 | Data Privacy in an AI World | 2025-11-02 | Search-history-as-ad-profile demonstration; 3 privacy problems: inference explosion, agent cascade, synthetic data; Stanford HAI 3 solutions |
| 19 | When Invisible Algorithms Judge You | 2025-11-05 | Harper v. Sirius XM/iCIMS + dynamic flight pricing; 3 frameworks (civil rights / equity / human rights); UN GP on Business and Human Rights |
| 20 | The EU AI Act's Reality Check | 2025-11-09 | Digital Omnibus simplification package; 4-month confession; risk pyramid; 3 paths to high-risk; child-abuse-detection deployment dilemma |
| 21 | Understanding How the EU Regulates AI | 2025-11-11 | 8-framework healthcare-startup compliance; foundation/horizontal-floors/vertical-pillars architecture; 5-report incident problem |
| 22 | U.S. AI Policy Myths and Realities | 2025-11-17 | Trump AI Action Plan 3 pillars; 4 myths (deregulation / comprehensiveness / one-voice / industry-unanimity); Anthropic-Sacks conflict |
| 23 | When Silicon Valley's EAs Meet Washington's Export Controls | 2025-11-18 | Yudkowsky p(doom); ITAR vs EAR; FDPR rule; Oct 7 2022 chip controls; AI Diffusion Rule; Huawei/DeepSeek workarounds; 3 scenarios |
| 25a | China's AI Governance | 2025-11-30 | Trump vs. Xi same-week press conferences; 3 China countermoves (compensate/control inputs/redirect race); paradox of domestic control + global open-source; 3 theories of victory |
| 25b | The Executive Order That Could Kill State AI Laws | 2025-12-11 | Trump Dec 11 2025 federal preemption EO; Senate 99-1 rejection of moratorium; 5 mechanisms; Pike v. Bruce Church; coercion vs inducement |
| 26 | When AI Stops Advising and Starts Acting | 2025-12-08 | Fowler Operator $31 eggs; agent autonomy spectrum (Levels 1-5); CMU TheAgentCompany 24-30% success; Air Canada chatbot; principal-agent law applied to AI |
Each class has a corresponding source page in Wiki/sources/, all tagged source_class: foundational. Source files in Raw Sources/ are titled <Class title> (Inside my AI Law and Policy Class N).md (or Class XIII for class 13).
The introductory course is the originating source for several cross-cutting frameworks that have dedicated concept pages: Embodied Perception (Tacit Knowledge vs. Pattern Recognition) (pattern recognition vs. embodied/tacit knowledge, also in the Feb 2026 standalone essay); Fiduciary AI (chatbots as health/legal/financial authorities without fiduciary duty); Technological Convergence (AI / Neurotech / Biometrics / Agents) (siloed governance failure across AI/neurotech/agents/biometrics); Five Paradigms of AI Manipulation Governance (Krook's framework adopted by Farahany — harm/info/design/relationship/cognitive); Three Privacy Problems AI Creates (inference/cascade/synthetic); Agent Autonomy Spectrum (5 Levels) (Levels 1-5 from reactive to fully autonomous); Mathematical Impossibility of Perfect Fairness (equal-rates vs equal-accuracy vs equal-meaning trilemma); Five Levels of Meaningful Transparency (notification/explanation/interrogation/contestation/system-change); Liar's Dividend (Chesney-Citron framing, in which the deepfake era erodes trust in true evidence); Principal-Agent Problem Applied to AI (agency law applied to AI agents); and Three Theories of Victory (US / EU / China AI Governance) (US frontier dependency vs EU regulatory floor vs China multi-dimensional unavoidability).
Advanced course: "Inside My Advanced Topics in AI Law and Policy Class"
Farahany's follow-on Spring 2026 Duke Law course was serialized across 30 classes and organized around cognitive liberty. Its three-pillar curriculum covers the architecture of attention (Weeks 2–4), the architecture of cognition (Weeks 5–7), and the architecture of self-knowledge and judgment (Weeks 8–11), with each week typically holding three classes (Mon/Wed/Fri).
| Class | Title | Date | Anchor Case / Doctrine |
|---|---|---|---|
| 1 | From How AI Works to What AI Does | 2026-01-19 | Pivot from Fall course; introduces cognitive liberty framework |
| 1.2 | What AI Does To Your Thinking | 2026-01-21 | Cognitive offloading; Ward et al. 2017 brain drain; Haidt vs Orben/Przybylski |
| 1.3 | Protecting Autonomy in Law, Take 1 | 2026-01-23 | 4 legal vocabularies (consent/capacity/undue influence/competency); doctrinal map |
| 2 | Can You Pay Attention? | 2026-01-26 | Attention economy; Packingham v. NC; right to direct attention |
| 2.2 | The Attention Evidence Gap | 2026-01-28 | Haidt vs Orben/Przybylski empirical contest; subgroup effects |
| 2.3 | The Laws That Miss the Point | 2026-01-30 | Privacy/consumer-protection/competition/constitutional law each capture a slice |
| 3 | 20 Clicks to Cancel | 2026-02-02 | FTC v. Amazon Iliad Flow $25M; dark patterns vs fraud; Mathur 2019 taxonomy |
| 3.2 | Why Dark Patterns Work | 2026-02-04 | 5 cognitive mechanisms; Johnson & Goldstein organ-donor study; loss aversion |
| 4 | Everyone Described Harm | 2026-02-09 | Facebook Files; Surgeon General; Murthy v Missouri; NM v Meta; LA bellwether |
| 4.2 | The Shield — Section 230 | 2026-02-11 | Stratton Oakmont; Zeran; Lemmon v. Snap product-design carve-out |
| 4.3 | Is an Algorithm Speech? | 2026-02-13 | Tornillo / Turner / Zhang v Baidu; Moody v NetChoice; speech-certainty principle (Austin & Levy) |
| 5 | The Perfect Friend | 2026-02-16 | Sewell Setzer III + Character.AI Daenerys persona; companion-chatbot category |
| 5.2 | "Please Do, My Sweet King" | 2026-02-18 | Garcia v Character.AI complaint walkthrough; May 2025 motion to dismiss ruling |
| 5.3 | Two States, Two Bets | 2026-02-20 | WA SSB 5984 vs CA SB 243 — disclosure-design vs categorical-restriction theories |
| 6 | What the AI Thinks It Knows About You | 2026-02-23 | Mobley v Workday class certification; EEOC v iTutorGroup; inferred-portrait harm |
| 6.2 | The Law's Toolkit (and Its Blind Spots) | 2026-02-25 | Title VII/ADA/ADEA + EEOC + FCRA + NYC LL144 + IL AIVIA + CA FEHA + EU AI Act |
| 6.3 | Your Brain Everywhere | 2026-02-27 | Inferred-portrait AI across employment/insurance/lending/education/healthcare/LE |
| 7 | The Persuasion Exchange | 2026-03-02 | FTC v Epic $245M; Coffee v Google; personalization gradient (5 levels) |
| 7.2 | The Statute That Couldn't Stretch | 2026-03-04 | Facebook v Duguid (2021); TCPA; S. 1629 loot box bill |
| 7.3 | Aristotle's Algorithm | 2026-03-06 | Aristotelian rhetoric (ethos/pathos/logos) applied to AI persuasion; see Inside My Advanced Topics Class 7.3: Aristotle's Algorithm (Farahany, March 2026) |
| 8.1 | The Biggest Lie on the Internet | 2026-03-16 | "I have read and agree" exercise; GDPR Article 7; consent failure |
| 8.2 | The Environment Is the Argument | 2026-03-19 | Fortnite V-Bucks; FTC dark patterns report; 5 categories of architectural manipulation |
| 8.3 | What the Law Is Trying to Do About It | 2026-03-22 | DELETE Act / COPPA / CAADCA mapped to 3 theories of consent failure |
| 9.1 | The Senate Just Agreed On Something | 2026-03-22 | COPPA 2.0 unanimous Senate passage; 3 regulatory strategies (access/design/data) |
| 9.2 | The Law That Kept Getting Blocked | 2026-03-24 | NM v Meta + LA verdict; UK Children's Code; case AGAINST new mandates (fraud / product liability / transparency) |
| 9.3 | The Ninth Circuit Told California How to Fix the Law | 2026-03-26 | NetChoice v Bonta II (Mar 12, 2026); doctrinal through-line |
| 10.1 | The Government Can Take Your Blood. Can It Take Your Thoughts? | 2026-03-29 | Schmerber / Fisher / Hubbell; foregone-conclusion doctrine |
| 10.2 | Two Courts, One Test, One Thumb | 2026-03-31 | Boucher / Comm v Jones / Payne (9th Cir) / Brown (DC Cir); Farahany cognitive evidence spectrum |
| 10.3 | When the Interface Is Neural | 2026-04-03 | Meta EMG band; cognitive-exertion paradox; see Inside My Advanced Topics Class 10.3: When the Interface Is Neural (Farahany, April 2026) |
| 11.1 | The Wall That Cuts Both Ways, and Who Speaks | 2026-04-06 | State action doctrine; Knight / Lindke v Freed; Zhang v Baidu; speech-certainty |
The Advanced Topics course is the originating source for additional cross-cutting frameworks that have dedicated concept pages: Cognitive Liberty (Farahany's organizing concept, the umbrella for the entire course); Farahany Cognitive Evidence Spectrum (4 Categories) (a 4-category replacement for the testimonial/physical binary — identifying / automatic / memorialized / uttered); Foregone Conclusion Doctrine (a limit on Hubbell's act-of-production privilege); Content vs Architecture Theory of Social Media Harm (competing diagnoses of social-media harm, which determine what counts as a "less restrictive alternative"); Three Theories of Consent Failure (Information / Capacity / Design) (an information / capacity / design diagnosis-and-remedy framework); and Algorithmic Speech Doctrine (an unresolved First Amendment question shaping every architectural-regulation strategy).
Relationships
- related: Luiza Jarovsky — both critique AI's encroachment on human autonomy from different vantages.
- depends-on: Cognitive Liberty (her central organizing concept).
- supports: Parasitic AI / Spiral Personas — Farahany's neural-data warnings predate Lopez's Spiral framing but share the dependency-and-autonomy concern.
- instance-of: Incriminating Thoughts Spectrum (her originated framework).