AI and the ADA: Disability Risk in Hiring Tools
AI ADA hiring risk arises when automated employment tools screen, score, or interview candidates in ways that disadvantage people with disabilities, exposing employers to liability under the Americans with Disabilities Act (ADA). The ADA applies to AI-driven hiring even when no human intends to discriminate. An employer is responsible for disability bias produced by a resume screener, video interview analyzer, gamified assessment, or chatbot it deploys. The two highest-risk failures are screening out qualified disabled applicants and failing to provide a reasonable accommodation during an automated assessment.
This article explains where ADA liability attaches in AI hiring systems, which laws and enforcement actions already apply, and the concrete controls that reduce exposure for employers and the vendors who sell these tools.
What does the ADA require of AI hiring tools?
The ADA prohibits employment discrimination against qualified individuals with disabilities and requires employers to provide reasonable accommodations unless doing so causes undue hardship. When an automated tool sits between an applicant and a hiring decision, three obligations carry directly into the software.
No screen-out. A tool may not eliminate a candidate who could perform the job's essential functions with or without a reasonable accommodation. An assessment that measures something other than an essential function, and that a disability affects, can produce an unlawful screen-out.
Reasonable accommodation. Applicants must be able to request an alternative format or process. A timed cognitive game, a one-way video interview, or a voice-based chatbot can each disadvantage someone with a motor, speech, vision, hearing, or cognitive disability who is fully qualified for the role.
No improper disability-related inquiry or medical exam. A tool that infers health conditions, mood, or neurological traits from facial movement, keystrokes, or voice can cross into a prohibited pre-offer medical examination.
The EEOC's technical guidance on AI under Title VII and the ADA states that an employer generally remains responsible under the ADA when it uses an algorithmic decision tool, including tools built and administered by an outside vendor. Delegation to a vendor does not transfer the legal duty.
Who is liable when a vendor builds the tool?
The employer that uses the tool to make hiring decisions holds primary ADA responsibility. Vendor exposure is also expanding. In Mobley v. Workday, a federal court allowed discrimination claims to proceed against Workday on the theory that an AI screening vendor can function as an agent of the employers using its software, rather than a neutral software seller. The plaintiff, a Black applicant over 40 who is disabled, alleged bias on the basis of age, race, and disability across more than 100 job applications. The ruling signals that "we only licensed the software" is not a settled defense.
Two practical points follow:
Employers should treat AI procurement as a compliance decision, not only an IT or HR-tech purchase. Contracts should require bias testing, accommodation support, and documentation.
Vendors should expect to be named directly and should build accommodation pathways and audit logs into the product rather than disclaiming responsibility in terms of service.
How do AI hiring tools discriminate against disabled candidates?
Disability discrimination in hiring AI is usually a side effect of measuring traits correlated with disability, not an explicit rule. The mechanism differs from race or sex bias because disability is heterogeneous: two qualified candidates can be screened out by entirely different tool behaviors.
Tool type: Resume screener
How it can disadvantage disabled candidates: May penalize employment gaps related to disability or medical leave and disadvantage candidates with nontraditional career paths.
ADA concern: Unlawful screening out of qualified applicants.
Tool type: One-way video interview
How it can disadvantage disabled candidates: May score facial expressions, eye contact, or speech fluency that are affected by autism, facial paralysis, stuttering, or blindness.
ADA concern: Screening out qualified candidates and potentially making disability-related inferences.
Tool type: Gamified or timed assessment
How it can disadvantage disabled candidates: May emphasize reaction speed or working memory that is unrelated to essential job functions, disadvantaging individuals with ADHD, motor impairments, or cognitive disabilities.
ADA concern: Screening out qualified candidates without providing reasonable accommodations.
Tool type: Voice or chatbot intake
How it can disadvantage disabled candidates: May misinterpret speech disabilities or fail to support AAC device users, while rigid input formats can prevent effective use by screen-reader users.
ADA concern: Screening out qualified candidates due to accessibility failures.
Tool type: Personality or affect analysis
How it can disadvantage disabled candidates: May infer mental health conditions or neurodivergence based on word choice, tone, or other behavioral signals.
ADA concern: Potentially constitutes a prohibited medical inquiry under the ADA.
The historical record shows how these failures surface in production. Amazon scrapped an internal AI recruiting tool, reported in 2018, after it learned to down-rank resumes associated with women. That outcome showed that a model trained on biased historical hiring data reproduces and amplifies that bias. While that case centered on sex, the same training-data dynamic applies to disability: if past hiring favored candidates without employment gaps or with conventional communication styles, the model learns those preferences as proxies.
Why is disability bias harder to audit than other bias?
Most bias-audit methods, including the statistical approach behind New York City's Local Law 144, compare selection rates across groups. Disability resists that method for three reasons:
Low and uncertain disclosure. Many applicants do not disclose a disability, so an employer often cannot construct a clean comparison group.
No single protected class. The ADA covers a wide and individualized set of conditions, so an aggregate "disability vs. non-disability" pass rate hides condition-specific harm.
Accommodation is individual. The legal question is often whether one person could have performed the job with an accommodation, which a group statistic does not answer.
ADA compliance for AI hiring therefore depends more on tool design, accommodation process, and validation against essential functions than on a single disparate-impact ratio.
Which laws govern AI in hiring beyond the ADA?
Several federal and state regimes now apply to automated employment decision tools. Employers using AI in hiring should map their obligations across this stack rather than treating the ADA in isolation.
Law or action: Americans with Disabilities Act (ADA)
Scope: Disability discrimination in employment.
What it requires or established: Prohibits disability-based screening, requires reasonable accommodations, and restricts improper medical inquiries during the hiring process.
Law or action: EEOC AI guidance
Scope: AI use under Title VII and the ADA.
What it requires or established: Employers remain responsible for employment decisions made with vendor-provided AI tools, and accommodation obligations continue to apply.
Law or action: NYC Local Law 144
Scope: Automated Employment Decision Tools (AEDTs) used in New York City hiring.
What it requires or established: Requires an annual independent bias audit, publication of the audit summary, and advance notice to candidates. Enforcement began in July 2023.
Law or action: Illinois AI Video Interview Act
Scope: AI systems that analyze recorded video interviews.
What it requires or established: Requires candidate notice, consent, and an explanation of how the AI evaluates video interviews. Effective January 2020.
Law or action: Colorado AI Act (SB 24-205)
Scope: High-risk AI systems, including those used in employment.
What it requires or established: Imposes a duty of reasonable care to prevent algorithmic discrimination and requires consumer disclosures.
Law or action: EU AI Act
Scope: AI systems used for hiring and employment that are listed in Annex III.
What it requires or established: Classifies these systems as high-risk, requiring risk management, data governance, human oversight, and other compliance obligations.
The EEOC has already enforced age-based screening through software. In the iTutorGroup matter, the company settled with the EEOC in 2023 for $365,000 after its application software automatically rejected older applicants. The same enforcement logic applies when a tool functions to exclude candidates based on disability: an automated rule that screens out a protected group is actionable regardless of the absence of human intent.
For a closer reading of how federal regulators frame employer responsibility for these tools, see the EEOC's AI hiring guidance and what it means for employers.
Does NYC Local Law 144 cover disability bias?
Local Law 144 requires an annual bias audit of automated employment decision tools, but its mandatory audit centers on sex and race-ethnicity categories and intersectional groups. It does not impose a full disability audit. Employers should not read a passing Local Law 144 audit as ADA compliance. The ADA duties around accommodation and screen-out exist independently and are not satisfied by a sex-and-race selection-rate report.
How can employers reduce ADA risk in AI hiring?
Reducing exposure requires controls at procurement, configuration, and operation. The following steps map to the ADA obligations above and to the documentation regulators look for.
Validate every tool against essential job functions. Confirm that what the tool measures relates to the actual essential functions of the role, documented in current job descriptions. Remove assessments that test traits a disability could affect without a job-related justification.
Build a visible, low-friction accommodation request path. Offer accommodation at the point of each automated step, not buried in a careers FAQ. Provide alternatives such as a live interview in place of a one-way video, extended time, or a different assessment format.
Audit for screen-out, not only selection-rate parity. Test whether candidates who request accommodations or use assistive technology can complete and pass each step. Include screen-reader users, speech-disability users, and motor-disability users in usability testing.
Restrict affect, emotion, and health inference. Disable features that score facial expression, tone, or personality in ways that infer a medical or psychological condition, which can constitute a prohibited inquiry or exam.
Demand documentation from vendors. Require bias-testing results, an accommodation feature description, accessibility conformance such as WCAG alignment, and a record of how scores map to job functions.
Keep a human in the decision. Ensure a qualified person can review and override automated rejections, especially where an accommodation was requested or assistive technology was used.
Log and retain records. Maintain audit logs of tool versions, configurations, accommodation requests, and outcomes, because regulators and litigants will ask for them.
What should AI hiring vendors build into their products?
Vendors carry growing exposure after Mobley v. Workday and should treat ADA support as a product requirement. Priorities include an in-product accommodation request workflow, alternative assessment formats, accessibility conformance testing, configurable removal of affect and emotion analysis, and exportable audit logs that an employer can hand to counsel or a regulator. Vendors that ship these controls reduce both their own liability and the liability of every employer that buys the product.
Next Steps Checklist
Action: Inventory every AI tool used in hiring decisions.
Owner: HR and Legal.
Output: A tool register documenting each vendor, the tool's function, and its data inputs.
Action: Map each AI tool to ADA requirements, EEOC guidance, and applicable state and EU regulations.
Owner: Legal.
Output: An obligation matrix.
Action: Validate each AI tool against documented essential job functions.
Owner: HR and an Industrial-Organizational (I/O) specialist.
Output: Evidence demonstrating job-relatedness.
Action: Add an accommodation request option at every automated decision point.
Owner: HR and the AI vendor.
Output: A functioning accommodation request process.
Action: Test AI systems with assistive technologies and users with disabilities.
Owner: Accessibility team and HR.
Output: Usability findings and disability screen-out assessment results.
Action: Disable affect, emotion, and health inference features.
Owner: HR and the AI vendor.
Output: Configuration records confirming the features are disabled.
Action: Require vendors to provide bias testing and accessibility documentation.
Owner: Procurement.
Output: Contract clauses and supporting compliance reports.
Action: Preserve audit logs for model versions, configurations, and decision outcomes.
Owner: IT and Legal.
Output: Records that satisfy audit and retention requirements.
FAQ
Does the ADA apply to AI hiring tools built by a third-party vendor?
Yes. Under EEOC guidance, the employer that uses an automated tool to make hiring decisions generally remains responsible under the ADA, even when an outside vendor built and runs the tool. Mobley v. Workday also allowed claims to proceed against a vendor directly, on the theory that an AI screening provider can act as the employer's agent. Both the employer and, increasingly, the vendor can face liability.
What is an ADA "screen-out" in automated hiring?
A screen-out happens when an AI tool eliminates a candidate who could perform the job's essential functions, with or without a reasonable accommodation, because of a disability. Common examples include a timed game that penalizes a motor disability or a video tool that scores speech fluency affected by a stutter. Screen-outs are unlawful even when the employer had no intent to discriminate against disabled applicants.
Is a bias audit under NYC Local Law 144 enough for ADA compliance?
No. Local Law 144's required audit centers on sex and race-ethnicity selection rates, not disability. The ADA imposes separate duties around reasonable accommodation, screen-out, and improper medical inquiries that a Local Law 144 audit does not address. Treat the two as distinct obligations, and do not assume a passing 144 audit means a tool is ADA-compliant.
Can AI video interview tools violate the ADA?
Yes. Video tools that score facial expression, eye contact, speech patterns, or tone can disadvantage candidates with autism, facial paralysis, speech disabilities, or blindness, producing a screen-out. Features that infer mood or psychological state may also constitute a prohibited disability-related inquiry. Employers using these tools should offer alternative interview formats and disable affect-analysis features that lack a job-related justification.
