Disqualified by Disability: The Exclusion of Disabled Workers by Digitized Hiring Assessments
Michal Luria, Matthew U. Scherer, Ariana Aboulafia, Dhanaraj Thakur · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3791485
Summary
Luria, Scherer, Aboulafia, and Thakur — researchers at the Center for Democracy and Technology — conduct a qualitative, human-centered study of how "digitized assessments" (personality tests, gamified cognitive tests, situational judgement tests, emotional intelligence tests, and AI-scored video interviews) affect U.S. disabled workers navigating hiring. The motivation is stark: the EEOC estimates 83% of employers (99% of Fortune 500) use some form of automated hiring tool, while the employment rate for disabled Americans is roughly half that of non-disabled peers. Seventeen participants — eleven hourly workers and six attorneys/law students, spanning cognitive, vision, hearing, ambulatory, self-care, independent-living disabilities, autism, brain injury, MS, epilepsy and chronic pain — completed up to eight simulated assessments on commercial "test-prep" platforms that mirror real employer tools, then participated in hour-long semi-structured interviews. The authors analysed the data with reflexive thematic analysis, explicitly grounding the study in the disability-rights principle "nothing about us without us" and including multiply marginalised disabled participants. Beyond surfacing individual accessibility failures, the paper interrogates the structural logic of digitized assessments: their historical roots in early-20th-century personality testing used to screen out "maladjusted" workers, their function as a "Computer Says No" deflection mechanism that diffuses employer accountability, and their tension with ADA protections and the EU AI Act.
Key findings
Findings cluster around four themes. (1) Inaccessibility: 9 of 17 participants (53%) could not complete at least one assessment, and 7 of 17 (41%) scored in the bottom 5th percentile on at least one test due to access barriers rather than ability. One platform offered only extended time (25% of assessments); the other offered no accommodations. Design choices — colour-only cues, high information density, timed graphical tasks (a risk-taking Balloon test, a Flanker attention task, Math Bubbles), and a Situational Judgment Test with 18 scenarios × 4 courses of action — systematically disadvantaged participants with low vision, colourblindness, cognitive disabilities, and those using screen readers. (2) Ineffectiveness: participants, many of them practicing attorneys, argued the tests measure test-taking ability, not job fit, and over-index on traits like "positivity" and "liveliness" that disadvantage autistic and mental-health candidates; in the AI video interview, facial-expression analysis classified only 1 of 14 participants as ever "happy," with 40% of all expressions flagged "neutral" and 18% "angry." (3) Cognitive and emotional tax: completing assessments produced shame, exhaustion, reduced self-worth, and "access labor" (the additional invisible work of making inaccessible tools usable). (4) Forced disclosure: participants felt that either failing or requesting accommodations would expose disability status before a conditional offer — conflicting with ADA protections. Several noted this creates a chilling effect that keeps disabled workers from applying at all.
Relevance
This paper is directly useful for accessibility practitioners advising HR, procurement, and legal teams. It names the specific assessment formats most likely to produce ADA-violating outcomes, provides evidence that WCAG conformance of assessment interfaces is necessary but insufficient, and gives concrete policy anchors: supplement (don't screen with) assessments, mandate human review, publish use of digitized assessments at the job-posting stage, and treat facial-expression analysis as inherently discriminatory against people with facial differences, autism, and motor impairments. The authors explicitly tie inaccessible assessments to the ADA's prohibition on pre-offer medical inquiries and to the EU AI Act's protections around automated decision-making — a useful legal framing for enterprise conversations. Limitations include the small U.S.-only sample, the use of test-prep rather than live employer platforms, and the qualitative design; the authors flag the need for international, disaggregated-by-disability work.
Tags: AI hiring · algorithmic hiring · digitized assessments · automated employment decision systems · AEDT · employment discrimination · gamified assessment · video interview · disability justice · access labor · accessibility barriers
Standards referenced: WCAG · ADA · EU AI Act