Author: Nita Farahany Source: https://nitafarahany.substack.com/p/what-the-ai-thinks-it-knows-about-you Published: February 23, 2026
A Substack essay by Nita Farahany published February 23, 2026, opening Week 6 (AI Personality Assessment / Inferred Profiles) of her Advanced Topics in AI Law and Policy course. The essay surveys AI-driven personality, mood, and risk assessment in hiring, insurance, lending, and education, and argues that the central harm has moved past disparate impact: AI systems now produce inferred cognitive portraits — personality, risk, propensity, and mood — that the subject never disclosed and cannot easily contest, yet that are used in material decisions.
Summary of argument
Farahany frames the class around the claim that algorithmic hiring and assessment harm extends beyond the classic problem of disparate impact. AI systems, she argues, generate inferred profiles of applicants and other subjects, then use those profiles to make consequential decisions about hiring, insurance, lending, and education. The essay distinguishes three categories of harm:
- Disparate impact — the classic ADA and Title VII problem (covered in Class 9 of the intro course).
- Inferred-profile harm — being judged on personality or mood characteristics inferred without consent and used for material decisions.
- Process opacity — applicants do not know what the system measured, on what basis, or how to contest it.
A recurring theme is the contestability gap. Traditional civil-rights enforcement assumes the affected individual can identify the discriminating decision and challenge it, but AI screening produces decisions whose reasoning is unavailable even to the deploying employer. Farahany notes that the EEOC's 2023 technical-assistance document attempts to address this, while how courts treat it remains open.
Key claims
The market for personality- and behavioral-assessment AI, per the essay, includes Workday, HireVue, Pymetrics, Plum, Modern Hire, Spark Hire, and Crystal Knows, and such tools have become standard at Fortune 500 hiring funnels. Farahany states that roughly 80% of large employers use some form of automated screening.
On what these systems infer, the essay lists Big Five personality traits (OCEAN), risk tolerance, conscientiousness, leadership potential, "culture fit," emotional state read from voice, face, or body language, and propensity for tenure or attrition. Most of these inferences, Farahany writes, are correlational rather than validated against job performance.
The class is anchored on two cases. In Mobley v. Workday (N.D. Cal., filed 2023), Derek Mobley, a Black applicant over 40 with anxiety, alleged that Workday's AI screening system rejected him from more than 100 jobs. The court certified a class action in May 2025, which Farahany describes as the first AI-discrimination case to reach class certification. She emphasizes that the defendant is the AI vendor rather than the employers, and characterizes this vendor-liability theory as the doctrinal core of the case. In EEOC v. iTutorGroup (2023), the EEOC reached a $365K settlement over algorithmic age discrimination in hiring, after the system rejected female applicants 55 and older and male applicants 60 and older; the essay identifies it as the first EEOC AI-discrimination settlement.
Farahany contrasts the US and EU regimes. The EU AI Act categorizes employment-context AI as "high risk" and bans emotion-recognition in workplaces and schools under Article 5(1)(f), while the US has no analogous federal rule.
Provenance
The essay is the opening installment of Week 6 of Farahany's Advanced Topics in AI Law and Policy course, published on her Substack on February 23, 2026.
Relationships
- part-of: Nita Farahany Advanced Topics course (Class 6 of ~30)
- related: Mathematical Impossibility of Perfect Fairness, Mobley v. Workday, Inc.
- previous: Inside My Advanced Topics Class 5.3: Two States, Two Bets (Farahany, February 2026) next: Inside My Advanced Topics Class 6.2: The Law's Toolkit (and Its Blind Spots) (Farahany, February 2026)