Agent Liability: When Employers Outsource Discrimination To AI Vendors

Artificial intelligence plays a growing role in employment decisions throughout California. Employers might use automated tools to screen job applications, rank candidates, track attendance, measure productivity, suggest promotions, or identify employees for termination. Vendors often promote these systems as objective, efficient, and free from human bias. Still, automated tools can repeat past discrimination, use inappropriate stand-ins for protected traits, or penalize actions related to disability or protected leave. California law usually holds employers responsible, even if an outside vendor provided or managed the system.

California Employment Law Applies To Automated Decisions

California’s Fair Employment and Housing Act prohibits covered employers from discriminating based on race, religious creed, color, national origin, ancestry, physical disability, mental disability, medical condition, genetic information, marital status, sex, gender, gender identity, gender expression, age, sexual orientation, reproductive health decision-making, military status, veteran status, and other protected characteristics. California Government Code Section 12940 applies to hiring, discharge, compensation, promotion, training, and other terms or conditions of employment.

California’s regulations governing automated-decision systems took effect on October 1, 2025. These systems use computer processes to make or help make decisions about employment benefits. They might use artificial intelligence, machine learning, algorithms, statistical models, or other data-processing methods. The rules make it clear that discrimination is still illegal if an automated tool harms someone based on a protected characteristic, even if a manager did not mean for that to happen.

Employers must review how an AI product actually works in their employment decisions. Simply trusting a vendor’s marketing, promises, or proprietary or undisclosed methods does not ensure compliance with California law.

Outsourcing A Decision Does Not Outsource Legal Responsibility

An AI vendor may act as an employer’s agent when the vendor performs recruiting, screening, assessment, ranking, or personnel-management functions on behalf of the employer. California Government Code Section 12926 includes an employer’s agent within relevant employment-law definitions. California Government Code Section 12940 also prohibits aiding, abetting, inciting, compelling, or coercing conduct forbidden by the Fair Employment and Housing Act.

Employers can be held responsible if a vendor’s system rejects applicants, gives lower scores, withholds opportunities, or suggests termination for illegal reasons. Vendors can also be liable if they knowingly take part in discrimination or do work that puts them under California employment laws.

A contract that says a vendor is responsible for legal compliance may shift financial risk between businesses. However, this does not always remove an applicant’s or employee’s right to make a legal claim against the employer. California civil-rights laws focus on the employment decision and any harm caused, not just on who created the software.

Disparate Impact Can Exist Without Intentional Bias

Automated discrimination does not always mean a system is directly programmed to consider race, disability, age, sex, or other protected traits. Even a facially neutral rule can be unlawful if it excludes members of a protected group at a higher rate than others.

For example, an AI screening tool may favor employment histories without gaps. That preference can disadvantage an applicant with a history of pregnancy disability leave, medical leave, military leave, caregiving leave, or another protected absence. A productivity system may penalize reduced output during an approved leave period. Attendance software may treat protected absences as misconduct. A promotion model may reward uninterrupted availability in a manner that disadvantages workers who exercised statutory leave rights.

Under California’s automated-decision regulations, an employment test, qualification standard, proxy, or other selection criterion that creates adverse impact may violate the Fair Employment and Housing Act unless the employer establishes that the practice is job-related and consistent with business necessity. A less discriminatory alternative may also undermine that defense.

Disability And Leave Data Create Serious Risks

AI systems can create particular problems for employees with disabilities or medical restrictions. Video-interview software may analyze facial movements, eye contact, speech patterns, response time, or vocal tone. Online assessments may measure reaction speed, memory, dexterity, or concentration. Such measurements can disadvantage an applicant with a physical, neurological, developmental, psychological, visual, hearing, or speech-related disability.

California Government Code Section 12940 requires reasonable accommodation for a known physical or mental disability unless accommodation would impose an undue hardship. The statute also requires a timely, good-faith interactive process. An employer cannot lawfully allow an automated process to block access to accommodation or reject an applicant because an assessment format measures disability-related traits rather than essential job functions.

Leave-related information presents similar concerns. California Government Code Section 12945 protects pregnancy-related rights, while California Government Code Section 12945.2 establishes qualifying leave protections under the California Family Rights Act. An algorithm that counts protected leave against an employee during a performance review, promotion decision, reduction in force, or termination analysis may support claims involving discrimination, interference, or retaliation.

Employer Knowledge And Vendor Oversight Matter

An employer should understand which data enters an AI system, which criteria influence an output, and how much authority a manager gives to an automated recommendation. A lack of technical knowledge does not necessarily excuse an unlawful employment action. Warning signs can include unexplained demographic disparities, unusually high rejection rates, repeated accommodation complaints, inconsistent scores, or vendor resistance to meaningful review.

California regulations recognize anti-bias testing and similar proactive measures as relevant evidence when evaluating automated discrimination. Regular testing should examine outcomes across protected groups, including applicants and employees with disabilities or leave histories. A valid review should also consider whether proxy variables closely correlate with protected characteristics.

Employment records connected to automated decisions may become important evidence. Such records can include input data, scoring criteria, model outputs, validation studies, audit reports, vendor communications, accommodation requests, manager notes, and final decision documents. California’s regulations expanded relevant recordkeeping obligations to cover automated-decision data for at least four years. 

Evidence In An AI Discrimination Claim

A worker may not know that an algorithm influenced an adverse decision. A rejection notice, sudden ranking change, unexplained performance score, or termination based on an unfamiliar metric may provide an early indication. Evidence can also arise from job postings, vendor disclosures, internal policies, emails, score reports, demographic patterns, or inconsistent explanations from management.

A discrimination claim may examine whether the employer relied on the system, whether the system caused a significant disadvantage, whether protected leave or disability information affected an output, and whether a qualified applicant or employee received less favorable treatment. Statistical evidence can become especially important in a disparate-impact case.

Potential remedies can include lost wages, lost benefits, emotional-distress damages, reinstatement, hiring, promotion, injunctive relief, attorney’s fees, and other relief permitted by law. Available remedies depend on the facts, procedural posture, responsible parties, and legal claims.

Frequently Asked Questions About AI Vendor Discrimination

Can An Employer Blame An AI Vendor For A Discriminatory Decision?

An employer generally cannot avoid liability merely by stating that a third-party vendor produced a score or recommendation. California employment law examines whether an unlawful practice affected hiring, promotion, compensation, discipline, leave, termination, or another employment benefit. An employer selects the technology, supplies or approves relevant data, determines how an output will be used, and ultimately controls the employment relationship. A vendor contract may provide indemnification or reimbursement rights between the businesses, but such a contract does not erase statutory protections held by an applicant or employee. Liability may extend to more than one entity when both businesses participated in the challenged process.

Can A Neutral Algorithm Still Violate California Law?

Yes. An algorithm can violate California law through disparate impact even when no protected characteristic appears as an express input. A system may use zip codes, employment gaps, school history, speech patterns, scheduling availability, attendance data, or other information that closely correlates with a protected characteristic. A neutral rule may disproportionately exclude older applicants, workers with disabilities, members of a racial group, pregnant employees, or employees with protected leave histories. An employer may need to prove that the challenged criterion is job-related and consistent with business necessity. The existence of a less discriminatory alternative can significantly affect that analysis.

How Can AI Discriminate Against An Employee Taking Protected Leave?

A system may count approved absences as attendance violations, reduce a productivity score during leave, treat an employment gap as a negative factor, or classify reduced availability as a lack of commitment. Such treatment can influence promotion, scheduling, compensation, performance ratings, layoffs, or termination. Protected leave cannot lawfully become a hidden penalty inside an automated scoring model. Relevant protections may arise under the California Family Rights Act, pregnancy disability provisions, disability-accommodation law, paid sick leave rules, or another applicable statute. The specific legal theory depends on the leave category, employer coverage, eligibility requirements, and employment action.

What Records May Help Prove Automated Discrimination?

Useful records can include application materials, assessment results, rejection notices, performance reports, attendance records, leave approvals, accommodation requests, disciplinary documents, and termination papers. Internal records may include algorithmic scores, model inputs, ranking reports, audit findings, validation studies, vendor instructions, manager communications, and demographic analyses. Comparative evidence concerning similarly situated workers may also help. A prompt preservation effort can be important because automated systems may update scores, overwrite prior results, or change model versions. California recordkeeping regulations require retention of significant employment and automated-decision information, but early action can still reduce the risk of missing evidence.

Is Intent Required For An AI Discrimination Claim?

Intent is not required for every claim. A disparate-treatment claim generally concerns intentional discrimination, although intent can be proven through circumstantial evidence. A disparate-impact claim focuses on the consequences of a facially neutral policy or selection practice. Statistical disparities, questionable proxies, inadequate validation, ignored audit results, or refusal to adopt a less discriminatory method may support such a claim. An employer’s failure to understand a purchased system does not automatically defeat liability. The legal inquiry centers on the challenged practice, the resulting disadvantage, and any available defense.

Can An AI Interview Tool Violate Disability Law?

Yes. A video or audio assessment may disadvantage an applicant based on speech, facial movement, eye contact, reaction time, hearing, vision, mobility, or neurological characteristics. An online test may also become inaccessible without additional time, screen-reader compatibility, keyboard access, captioning, or another adjustment. California Government Code Section 12940 requires reasonable accommodation and a good-faith interactive process in covered circumstances. An employer should provide a meaningful method for requesting accommodation and an alternative assessment when needed. Automatic rejection following a disability-related performance difference may support a discrimination claim when the measured trait is not essential to the position.

Can An AI Vendor Face Direct Liability?

A vendor may face direct liability depending on the vendor’s role, authority, knowledge, and participation. A vendor acting as an agent, employment agency, or covered entity may fall within provisions of California employment law. Liability may also arise when a vendor aids or assists an unlawful practice. A technical provider with no control over an employment decision may present a different legal analysis from a vendor that designs criteria, ranks candidates, recommends termination, or administers assessments. Contract terms, actual conduct, access to data, and decision-making authority can all affect responsibility.

What Should An Employee Do After Suspecting AI Discrimination?

An employee should preserve employment records, electronic notices, screenshots, score reports, leave documents, accommodation communications, and written explanations provided by management. A written chronology can identify dates, decision-makers, protected activity, changes in treatment, and statements concerning automated tools. Publicly available vendor materials may also describe relevant scoring methods. An employment attorney can assess possible claims, filing deadlines, responsible entities, and evidence-preservation needs. Delay can affect access to records and statutory rights, so prompt legal review is important.

Call Law Office Of Joseph Richards, P.C. For A Free Consultation

Automated employment systems can conceal discrimination behind scores, rankings, and vendor-generated recommendations. A California applicant or employee may have legal rights when an AI tool penalizes protected leave, disability, age, race, sex, medical condition, or another protected characteristic. Law Office of Joseph Richards, P.C. represents clients throughout the state of California in employment matters involving leave-of-absence violations and related discrimination.

If you or someone you know has been impacted by employment discrimination, please contact our Orange County employment discrimination lawyer at the Law Office of Joseph Richards, P.C., at 877-562-9637 or 877-JOB-WOES to receive your free consultation. A timely case review can help identify responsible parties, preserve automated-decision records, evaluate vendor involvement, and determine available remedies under California employment law.