AI hiring and disability discrimination: your rights at work

U.S. employment guide. Sources checked September 30, 2026.

In 2015, I was diagnosed with chronic myeloid leukemia. In 2023, I had open-heart surgery. Neither event came with instructions for making it look attractive on a CV.

I am a former United Nations humanitarian professional and a disabled person living with chronic illness. Those facts belong to the same person. When I read about software judging candidates by their employment history, speech or facial movements, I wonder how much of that person it could possibly understand.

A hiring system can produce a score with several decimal places. That does not mean it knows what it is talking about.

AI hiring tools can discriminate against disabled applicants when their design or use creates unlawful barriers. The Americans with Disabilities Act still applies: covered employers must assess people fairly and provide reasonable accommodations where required. Buying the software from someone else does not remove that responsibility.

If you have an assessment invitation sitting in your inbox, go straight to the accommodation request. You do not need to become an expert in machine learning before asking for an interview you can actually take part in.

In this guide

• Where automated hiring goes wrong
• What the ADA requires
• Requesting an accommodation, with an email example
• What to do if a request is refused
• AI after you get the job
• A practical employer checklist
• Illinois video-interview rules and other questions

How AI hiring tools can discriminate

An automated hiring system might filter resumes, rank applicants or help score an assessment. Some tools use machine learning; others follow rules written by people. A recorded interview may simply be a recording for a recruiter to watch. Do not assume that every video platform analyzes faces, or that every rejection came from AI. Ask what the employer actually uses.

The problem often starts with what the system treats as evidence of a good employee.

An uninterrupted employment history may look reassuring to a recruiter. It says much less about the abilities of someone who needed cancer treatment. Eye contact might be easy to measure on camera. It is a poor substitute for finding out whether someone can do the work.

Some predictive systems learn from past hiring or employee data. If that history reflects exclusion, copying it more efficiently can reproduce the same exclusion. Other barriers need no sophisticated algorithm at all: a test that does not work with a screen reader can stop a qualified candidate just as effectively.

Resumes and disability-related experience

A University of Washington study presented in 2024 tested GPT-4 by comparing a resume with versions containing additional disability-related credentials. Across 60 comparisons, the versions with those credentials ranked first only a quarter of the time. Some explanations drew on disability stereotypes.

That was a controlled experiment involving one model and a particular task. It does not establish how every recruitment product behaves today. It does show why an employer should demand evidence before trusting a fluent explanation of a candidate’s supposed shortcomings.

A separate 2025 study involved 528 participants choosing candidates with simulated AI recommendations. Participants tended to follow the system’s racial biases. This was research on racial bias, not a follow-up test of disability discrimination. Its relevance here is narrower: putting a person after an algorithm does not, by itself, make the decision fair.

Interviews, tests and the wrong things being measured

The following are possible barriers, not a claim that every platform has these features or that everyone with a particular disability needs the same adjustment.

The American Foundation for the Blind’s June 2026 employment report gives these concerns substance. Participants described assessments that interfered with accessibility software and a video setup task requiring movements a physically disabled applicant could not complete independently.

The report drew on survey responses, including 362 people who had recently looked for work. It examined automated assessments broadly because respondents could not always tell whether a system used AI. These accounts document access problems; they do not establish that every reported incident was an ADA violation.

Disabled advocate and attorney Lydia X. Z. Brown put the wider issue plainly during a public Sight Tech Global discussion of AI and bias:

“[Algorithmic discrimination] builds on the ableism and other forms of oppression that have already existed throughout society.”

— Lydia X. Z. Brown

The software did not invent the idea that a competent person must speak or move a particular way. Someone taught it what to reward.

What the ADA requires when employers use AI

Under Title I of the ADA, covered employers generally include private employers and state or local government employers with at least 15 employees. For smaller public agencies, DOJ’s Title II regulations also address employment discrimination, although courts differ on whether employees can bring Title II claims. Do not assume that having fewer than 15 employees means an agency has no duties; get advice about the protections and filing route that apply. State laws may also cover smaller employers. Federal applicants and employees have protections under the Rehabilitation Act and a different complaints process.

A qualified applicant must meet legitimate job requirements and be able to perform the job’s essential functions, with or without reasonable accommodation. That distinction matters. Access to an assessment does not mean being excused from the work the job genuinely requires.

A vendor’s product is still the employer’s responsibility

The Department of Justice’s guidance on AI and disability discrimination in hiring explains that employers can be responsible for discriminatory hiring technology supplied by another company.

An employer needs to understand what its system measures and how disabled applicants can participate. “The platform won’t let us” is a reason to investigate the platform, not the end of the accommodation conversation.

The legal issues include:

• Unlawful screening out. A selection standard that excludes disabled applicants may need to be justified as job-related and consistent with business necessity. Required accommodations still matter.
• Failure to accommodate. Employers must provide reasonable accommodations for known disability-related needs unless doing so would cause undue hardship, meaning significant difficulty or expense.
• Improper medical inquiries. Before a conditional offer, employers generally cannot ask disability-related questions or require medical examinations, subject to limited exceptions. A digital assessment does not escape those rules.

These duties come from the law. An accessibility badge or a vendor’s bias audit cannot replace them.

Archived guidance does not repeal the ADA

The DOJ’s broader Artificial Intelligence and Civil Rights page is now archived and warns that its contents may be outdated. Its separate ADA.gov hiring guidance and downloadable PDF were still accessible when checked on September 30, 2026.

The EEOC’s AI resource index was also accessible, but its link to the main 2022 technical-assistance document returned a “page not found” response. This guide therefore links directly to the ADA statute, live DOJ guidance and the EEOC’s Job Applicants and the ADA guidance.

Guidance explains agency interpretations; it can change or disappear. The ADA’s employment protections remain in force.

How to request an accommodation for an AI interview

Tell the recruiter, HR team or named accommodation contact what part of the process creates a disability-related barrier and what change would help. Ask as soon as you know you need it, ideally before starting the assessment.

The EEOC says a request can be spoken or written. You do not need a legal phrase or a particular form of words. Writing is useful because it gives both sides a record.

You can begin by describing the relevant limitation rather than sending your medical history. If your disability or need for accommodation is not obvious, the employer may request reasonable documentation establishing the disability and why the adjustment is needed. That is different from an unrestricted right to your medical records.

Explain the barrier, then propose a change

For example:

• “My speech disability affects automated transcription. Could we arrange a live interview with an accessible communication format?”
• “I use a screen reader, but this assessment blocks it. I need a compatible version or an alternative assessment.”
• “My disability affects how quickly I use a mouse. Could I have additional time, or demonstrate the relevant skill another way?”
• “I need disability-related breaks during the test. Please confirm how these will be accommodated and how the proctor will be informed.”

Other possibilities include captions, an interpreter, written instructions or advance questions where appropriate. The adjustment must address your particular barrier and allow a meaningful assessment of the relevant skills.

You can request a human interviewer. You are not automatically entitled to your preferred format if the employer offers another effective accommodation. If the alternative does not work, explain why. “We have provided an alternative” is not enough if you still cannot use it.

For ideas, the Job Accommodation Network provides free, confidential accommodation guidance.

Accommodation request email

The example below requests a live interview for a speech-related barrier. Adapt the details to your situation.

Subject: Interview accommodation request

Hello,

Thank you for inviting me to the next stage of recruitment. I have a disability that affects my speech, and I am concerned that the automated video assessment will not accurately evaluate my answers.

I am requesting a reasonable accommodation: a live interview with a member of the hiring team, using an accessible communication format. If you propose another approach, I would be happy to discuss whether it addresses the barrier.

Please confirm who will handle this request and whether you can extend the assessment deadline while we agree on arrangements. I would appreciate written confirmation before I begin.

Thank you.

A request does not automatically pause an assessment deadline. Ask for confirmation and keep the reply.

If your request is refused or the system screens you out

A rejection alone does not prove discrimination. It is still worth documenting an inaccessible assessment, a refused adjustment or a score that appears to reflect your disability rather than your ability to do the job.

Keep the job advertisement, assessment instructions, relevant screenshots, emails and dates. Note which software feature caused the problem and how. Do this while the details are fresh and the application portal remains accessible.

Ask the employer for its reason in writing, whether an effective alternative is available, and whether someone can review the decision using accessible evidence of your skills. You do not have to wait for that conversation to finish before seeking advice.

EEOC filing deadlines

For most private-sector and state or local government employment claims, an EEOC charge must generally be filed within 180 calendar days of the discriminatory act. That period generally extends to 300 calendar days when a state or local agency enforces a law prohibiting employment discrimination on the same basis.

Federal employees and federal job applicants generally must contact an agency EEO counselor within 45 days, under a separate process. Exceptions can apply, so contact the relevant agency promptly.

Internal grievances or negotiations generally do not stop the EEOC filing clock. Start with the EEOC’s instructions for filing a charge. The EEOC explicitly distinguishes an online inquiry from a filed charge. An inquiry or interview appointment is not itself a signed charge; make sure the filing is completed in time.

This is general U.S. information, not advice on an individual legal claim. An employment lawyer or the relevant civil-rights agency can help assess your circumstances and deadline.

AI at work after you get the job

The same questions follow you into employment. What is the tool measuring? Does it account for accommodations? Who can correct a misleading result?

The AFB report includes accounts of monitoring software treating disability-related breaks as evidence of underperformance. It describes accommodation conflicts as uncommon: 12 disabled workers reported that an accommodation was incompatible with workplace surveillance. If your employer uses automated productivity measures, ask how approved breaks, assistive technology or a modified schedule are reflected in those measures. An accommodation should not become a mark against you when the monthly report arrives.

AI can also provide access. Workers in the same research used features such as captioning and visual description, though permissions and workplace policies sometimes obstructed them. Whether a tool works as an accommodation depends on its accuracy, accessibility and the person’s needs. Ask the employer to assess an effective, secure arrangement rather than assuming that an approved AI product will automatically be usable by everyone.

What employers should change

An employer can start by explaining the recruitment process before the applicant opens the test. Who reviews the answers? Does software score them? How does someone reach a person if the format is inaccessible?

Those are reasonable questions. They should not require detective work.

The following are practical recommendations. They combine ways to support legal compliance with design choices that go beyond a universal legal minimum:

• Test the whole application journey with disabled users. Include keyboard navigation, screen readers, magnification and captions. Use WCAG 2.2 AA as a design target; conformance alone does not establish that hiring decisions comply with the ADA.
• Give candidates a clear accommodation route before the assessment. Name a contact, explain the process and make sure someone can act before the deadline expires.
• Provide effective alternatives. Have an accessible assessment or interview option ready, and make sure requesting it does not reduce a candidate’s prospects.
• Require disability-specific evidence from vendors. Ask which disabilities, assistive technologies and actual tasks were tested. A race-and-sex audit does not answer those questions.
• Assess skills the job needs. Challenge unexplained scores for facial expression, voice or “culture fit.” Avoid facial or emotion analysis for hiring judgments; do not treat a vendor’s label as evidence of validity.
• Give human reviewers authority to question the score. They need training, relevant evidence and the ability to correct a decision, rather than a screen that encourages them to approve it.
• Keep reviewing the system. Check accommodation problems and outcomes after changes to the software or recruitment process, while protecting applicants’ confidential information.

Clear inclusive job descriptions help establish what the role actually requires before anyone builds a scoring system around it.

For the interview itself, URevolution’s guide to interviewing autistic candidates discusses concrete ways to reduce communication and sensory barriers. Ask the individual what works for them. “Neuroinclusive” means very little if the applicant cannot get an answer about extra time.

Questions about AI interviews and disability

Do I have to disclose a diagnosis to every recruiter?

No. You generally do not have to volunteer disability information simply to apply for a job. To request an accommodation, explain that you need a change because of a disability or medical condition. An employer may seek reasonable supporting documentation when the disability or need is not obvious. You do not usually need to begin by naming a diagnosis.

Can I refuse an AI interview?

You can raise an objection or request a disability-related accommodation. The ADA does not create a blanket right for every applicant to reject all automated assessment. Explain the access barrier and discuss an effective alternative. Separate state or local rules may also apply.

Does Illinois require consent for AI video interviews?

Yes, within the scope of its Artificial Intelligence Video Interview Act, 820 ILCS 42. For positions based in Illinois, an employer asking applicants to submit recorded interviews for AI analysis must give advance notice, explain how the AI works and the general characteristics it evaluates, and obtain consent before the interview. It cannot use AI to evaluate an applicant who has not consented.

The Act also limits sharing of the videos and requires deletion within 30 days of an applicant’s request, including instructions to recipients to delete their copies. It does not itself promise every applicant a live replacement interview.

Illinois also has broader AI employment provisions in its Human Rights Act, effective January 1, 2026, addressing discriminatory effects and notice. Those are separate from the video-interview consent law. This guide is not a complete account of Illinois’ implementation rules.

Are remote-proctored certification exams covered by the same rules?

Not necessarily. A test used by an employer to select applicants and an exam administered by a licensing or certification body can fall under different legal provisions. This guide focuses on employment. For a separate credentialing exam, contact the testing body about its accommodation process rather than assuming the employer rules and EEOC deadlines apply.

Does a fast rejection prove an algorithm discriminated against me?

No. A quick rejection does not reveal who made the decision or why. Look for evidence of the actual barrier: inaccessible controls, disability-related scoring, instructions that conflict with assistive technology or an accommodation refusal. Keep the record and seek advice.

Employers choose what counts as evidence of ability. If their technology cannot distinguish a disability-related access need from an inability to do the job, they need to change the assessment. The applicant has already applied for one job. Fixing the recruitment system should not become a second, unpaid one.