Artificial intelligence is becoming a routine part of hiring. Employers use automated tools to screen resumes, rank candidates, administer assessments, schedule interviews, transcribe or evaluate interview responses, and identify applicants for further review. Those tools may improve speed and consistency, but they also create employment-law risk when the employer cannot explain how the tool works, what information it uses, or why an applicant was screened out. The federal enforcement environment is changing, but that change does not eliminate exposure under federal statutes, private litigation, state law, or local AI-specific rules.
Employers should keep the following six issues in mind before relying on AI or other automated tools in the hiring process.
1) Federal Policy Has Changed, but Discrimination Claims Have Not Disappeared.
In April 2025, Executive Order 14281 directed federal agencies to reduce reliance on disparate-impact theories of discrimination. Disparate impact refers to a facially neutral workplace practice that disproportionately harms members of a protected group and cannot be justified under the applicable legal standard. Subsequent federal activity has further signaled reduced federal emphasis on disparate-impact enforcement. Even so, employers should not underestimate the risk of disparate impact. The underlying federal antidiscrimination statutes remain important, private plaintiffs may continue to assert claims, courts will determine how those claims proceed, and state or local agencies may continue enforcing their own civil-rights laws.
Washington employers also face exposure under the Washington Law Against Discrimination and related local ordinances. A neutral hiring practice that screens out applicants in a way that disproportionately affects a protected group can still create risk, particularly if the employer cannot show that the practice is job-related, consistent with business necessity, and supported by reliable validation or review.
For employers using automated screening tools, the practical question is far beyond simply whether anyone intended to discriminate. Employers should also ask whether the tool produces materially different outcomes for different groups of applicants, whether those outcomes are tied to legitimate job requirements, and whether a less exclusionary alternative is available.
2) Using a Vendor Does Not Transfer the Risk.
Many employers rely on outside vendors for resume screening, candidate ranking, assessments, and other automated hiring functions. That arrangement does not necessarily insulate the employer from liability.
For employers, vendor selection should therefore involve more than comparing features and price. Before adopting a tool, employers should understand:
- What applicant data, proxies, assumptions, and criteria the system uses;
- How candidates are scored, ranked, filtered, or recommended;
- Whether the vendor performs bias testing, validation, and monitoring after model or configuration changes;
- Whether the employer can obtain audit results, validation materials, decision logs, and other information needed to defend a challenged hiring decision;
- Who controls relevant data and how long records will be retained; and
- How the contract allocates responsibility for notice, accommodations, bias testing, indemnity, defense costs, and litigation cooperation.
“The vendor built it” is unlikely to be a satisfactory response when an employer is asked to explain why a candidate was rejected or why a selection process produced unequal outcomes.
3) Remote Hiring Can Create Compliance Obligations Outside Your Home State.
An employer’s legal obligations do not necessarily stop at the state line. States and cities have adopted, and continue to consider, different rules governing automated employment decision tools. Depending on the jurisdiction and the particular employment arrangement, employers may face requirements involving applicant notice, bias audits, impact assessments, recordkeeping, human review, or restrictions on discriminatory AI use. This is especially important for remote hiring. A Washington-based employer may recruit applicants or fill positions connected to jurisdictions with very different AI rules.
The practical lesson is to evaluate compliance based on the employer’s actual hiring footprint, not headquarters alone. Before using an AI hiring tool, employers should identify where they recruit, where applicants reside, where employees will work, and whether any of those jurisdictions impose additional notice, audit, documentation, or process requirements.
4) Disability and Accommodation Issues May Be the Most Overlooked Risk.
Some of the greatest risks from automated hiring tools involve applicants with disabilities. A tool may measure traits that overlap with a disability even when disability is not the stated subject of the assessment. Risk can arise when the employer fails to provide notice, an accessible process, or a meaningful opportunity to request an accommodation before the tool screens out the applicant. For example:
- timed assessments may disadvantage applicants whose disabilities affect processing speed, concentration, or motor function;
- video-analysis tools may interpret and evaluate eye contact, facial expression, speech patterns, or movement in ways that disadvantage some applicants;
- communication or adaptability assessments may affect neurodivergent candidates differently; and
- automated interview systems may reject a candidate before the candidate ever has an opportunity to request an accommodation.
These issues can create problems under the Americans with Disabilities Act and state disability-discrimination laws, including where a neutral tool operates as an unlawful screening device or where the employer fails to provide a reasonable accommodation.
Employers should make accommodation information visible before an applicant enters an automated assessment process. They should also provide a practical way for candidates to request an alternative process, ensure that requests are routed to a person who can respond promptly, and build in human review before an automated tool produces an adverse result that may have been affected by a disability-related limitation.
5) Testing and Recordkeeping Matter Before a Claim Is Filed.
Bias testing, validation, and recordkeeping matter should occur before it’s too late. Testing can help an employer identify problems, evaluate whether a tool is job-related, and show that the employer exercised reasonable diligence before relying on automation. But testing can also create documents, data, and communications that become important in later litigation.
Employers should work with counsel when appropriate to determine the purpose, scope, methodology, and documentation of any bias audit or validation review before testing begins. Simply involving an attorney does not automatically make testing confidential, and privilege issues should be considered at the outset rather than after a dispute arises.
6) Recordkeeping Presents a Related Problem.
AI systems change. Vendors update models, employers adjust settings, job criteria evolve, and applicant data may be stored in multiple systems. If an applicant challenges a decision months or years later, the employer may need to explain what version of the system was used, what information went into it, what result it produced, whether an accommodation was requested, and whether a person reviewed the decision. If those records no longer exist, defending the hiring decision becomes significantly harder.
What You Should Do Now
Employers do not necessarily need to stop using AI in hiring, but they should know where it is being used, what it does, and how decisions are made. A useful starting point is to:
- Inventory hiring technology. Include tools that score, rank, filter, match, recommend, transcribe, assess, or communicate with candidates—not just the primary applicant tracking system.
- Map the hiring footprint. Identify where applicants reside, where recruiting occurs, and where selected employees will work.
- Review vendor agreements. Address bias testing, validation, model changes, access to decision data, record retention, litigation cooperation, liability, and indemnification.
- Create an accommodation process. Give applicants clear notice and a practical way to request assistance or an alternative assessment before an automated process screens them out.
- Keep humans involved. Important hiring decisions should be explainable and subject to meaningful human review, not merely the product of an unexplained automated score.
- Preserve records. Retain enough information to reconstruct how the system operated, what version was used, what data was considered, what result was generated, and who reviewed the decision.
AI hiring tools can make recruiting faster and more efficient, but automation does not automate away the employer’s legal responsibilities. Employers that understand their tools, document their processes, preserve meaningful human oversight, and plan for accommodations will be better positioned as this area of employment law continues to develop.
Please contact Bill Symmes or another member of the Williams Kastner Labor & Employment practice team for assistance in evaluating and managing the legal risks associated with using AI in the hiring process.
Authored by Bill Symmes and Kai Zhao.