AI in Hiring: What California and Texas Employers Must Do Now

If your company uses software to screen resumes, rank candidates, score interviews, recommend applicants, or decide who moves to the next round, you may already be using what regulators call an automated decision system or automated decisionmaking technology.

That label matters because AI in hiring is no longer a future-risk issue. It is a present compliance issue.

California and Texas have both adopted legal regimes that can reach AI or automated systems used in employment decisions — but they took very different approaches. California focuses heavily on whether the tool causes discriminatory effects. Texas focuses more narrowly on intentional discrimination. Federal law still applies in both places.

For employers hiring across state lines — especially startups and small-to-midsize businesses using applicant-tracking systems, assessment platforms, video-interview tools, or vendor-built screening software — the practical question is simple:

Do you know what your hiring tools are doing, and can you prove they are lawful?

If the answer is “not really,” now is the time to fix that.

First, What Counts as “AI in Hiring”?

Most employers hear “AI in hiring” and think of futuristic tools that make final hiring decisions without human involvement. That is not how regulators usually look at it.

California’s employment regulations define an automated-decision system broadly. In practical terms, an ADS can include a computational process that makes or facilitates an employment decision, whether the tool relies on artificial intelligence, machine learning, algorithms, statistics, or other data processing. The label on the software does not control. What the tool does is what matters.

Common examples can include tools that:

• screen resumes for particular terms, credentials, patterns, or gaps;
• rank or score applicants;
• recommend which applicants should move forward;
• score video interviews;
• analyze facial expression, speech, voice, or behavior;
• administer games, puzzles, personality assessments, or cognitive tests;
• target job advertisements to particular audiences;
• recommend candidates for hiring, promotion, discipline, or termination; or
• help managers decide who should receive an interview, offer, raise, promotion, or termination.

Two points surprise many employers.

First, a tool can matter legally even if a human makes the final decision. If the system influences who gets seen, ranked, interviewed, rejected, promoted, or disciplined, it can still shape the employment decision.

Second, ordinary workplace software is not automatically a regulated hiring tool just because it uses data. A spreadsheet, calendar, calculator, word processor, or ordinary database is not usually the issue unless it is being used to make or facilitate an employment decision.

The question is not “Did we buy an AI product?” The question is:

Is technology shaping who gets hired, promoted, disciplined, or screened out?

If yes, the employer should treat the tool as a compliance issue.

California: The Effects-Based Approach

California has taken one of the most aggressive approaches in the country.

The California Civil Rights Council adopted employment regulations addressing automated-decision systems under the Fair Employment and Housing Act. Those regulations were approved and filed in June 2025 and became effective October 1, 2025. The California Civil Rights Department’s rulemaking page identifies the regulations as the Employment Regulations Regarding Automated Decision Systems and confirms the October 1, 2025 effective date. California Civil Rights Department, Civil Rights Council Rulemaking Actions.

For employers, the core point is this:

California is not only asking whether an employer intended to discriminate. It is also asking whether the tool produced discriminatory effects.

That means a facially neutral tool can create liability if it screens out applicants or employees based on a protected characteristic, unless the employer can establish the legally required justification. In California, as under federal disparate-impact principles, good intentions are not enough.

A resume-screening tool may be designed to find “high performers.” A video-interview platform may be designed to predict “communication ability.” A game-based assessment may be designed to identify “problem solving.” But if the tool disproportionately screens out candidates in a protected group and the employer cannot justify the practice as job-related and consistent with business necessity, the employer may have a problem.

California’s rules also address the use of proxies — factors that may appear neutral but are closely correlated with protected traits. That matters because automated systems often rely on patterns that are not obviously discriminatory on their face. A model may not use race, sex, disability, age, or national origin directly. But it may use data points that function as substitutes.

Examples of potentially risky proxy variables can include:

• ZIP code or commute distance;
• graduation year;
• school attended;
• employment gaps;
• availability patterns;
• speech, facial, or behavioral data;
• prior salary history;
• credit-related indicators;
• social-media activity; or
• other data that correlates with protected characteristics.

Not every use of those variables is unlawful. But employers should understand exactly what the tool uses, why it uses it, and whether the tool’s outputs produce skewed results.

California Employers Cannot Outsource the Risk to Vendors

One of the biggest traps for employers is assuming that vendor-built software means vendor-owned risk.

That is not how employment-discrimination law usually works.

If an employer uses a third-party tool to help make hiring, promotion, discipline, or termination decisions, the employer may still be responsible for the result. A vendor may have designed the algorithm. A staffing agency may have administered the screen. An applicant-tracking platform may have generated the ranking. But the employer is the one using the result in an employment process.

That means employers should not accept vague vendor assurances like:

• “Our tool is compliant.”
• “The algorithm is bias-free.”
• “The platform uses objective data.”
• “The system has been validated.”
• “Other employers use it.”

Those statements are not enough.

Employers should ask vendors specific questions:

• What data does the tool use?
• What variables affect the score, rank, recommendation, or exclusion?
• Has the tool been validated for the specific job at issue?
• Has the tool been tested for disparate impact?
• When was testing last performed?
• What protected groups were included in testing?
• What were the results?
• What changes were made after testing?
• Can the vendor produce records if the employer receives a charge, lawsuit, agency inquiry, or candidate complaint?
• Does the contract require the vendor to cooperate with audits, investigations, litigation holds, and record requests?
• Does the vendor indemnify the employer for tool-related claims, regulatory inquiries, or failures to provide legally required documentation?

The practical rule is simple: if a vendor’s tool affects your hiring decision, you need enough information to defend your use of that tool.

Anti-Bias Testing Is No Longer Optional in Practice

California’s ADS employment regulations do not create a simple, universal rule that every employer must publish a bias audit before using a tool. But that does not mean employers can ignore testing.

Anti-bias testing is central evidence. Regulators, plaintiffs’ lawyers, and courts will want to know whether the employer evaluated the tool, what the testing showed, and what the employer did after learning the results.

A good testing process should address at least four questions:

1. What is the tool measuring? The employer should be able to explain what the tool claims to measure and why that measure is relevant to the job.
2. Is the measure job-related? The employer should connect the tool’s criteria to actual job duties, not vague ideas of “culture fit,” “executive presence,” “personality,” or “communication style.”
3. Does the tool create adverse impact? The employer should evaluate whether the tool disproportionately excludes or disadvantages candidates based on protected characteristics.
4. What did the employer do with the results? Testing is not useful if the employer ignores the outcome. If the tool produces skewed results, the employer should investigate, modify, suspend, replace, or narrow the tool as appropriate.

For many employers, anti-bias testing should be coordinated through counsel. That does not make every fact magically privileged, and privilege rules vary by context. But legal oversight can help structure the review, identify risk, and preserve appropriate confidentiality where available.

Bottom line: in California, not testing may become its own evidence problem.

Recordkeeping: Keep the Data You May Need Later

California employers should also prepare for longer and more detailed recordkeeping.

Employers using automated systems in employment decisions should preserve relevant records, including:

• job postings and selection criteria;
• applications and resumes;
• tool inputs;
• tool outputs, including scores, rankings, recommendations, flags, or exclusions;
• criteria or weighting used by the tool;
• validation materials;
• anti-bias testing results;
• vendor documentation;
• human-review notes;
• communications with vendors about tool performance;
• notices provided to applicants or employees; and
• policies governing the use of automated systems.

California’s ADS-related employment recordkeeping expectations are longer than many employers are used to. Employers hiring in California should plan around a four-year retention period for relevant employment and ADS-related records.

That can be difficult because some vendors do not automatically retain the underlying data. Some platforms overwrite scores. Some systems produce rankings but not explanations. Some vendors resist sharing model information. Some tools are configured differently for different jobs or locations.

Employers should solve those issues before a complaint is filed — not after.

California’s Second Layer: Privacy

California’s anti-discrimination rules are only one part of the picture.

California’s privacy framework adds another layer for automated decisionmaking technology. The California Privacy Protection Agency has finalized regulations addressing ADMT, including ADMT used for “significant decisions.” The CPPA has stated that businesses using ADMT to make significant decisions must comply beginning January 1, 2027, with related risk-assessment obligations subject to separate phase-in requirements. California Privacy Protection Agency, Regulations; CPPA, California Finalizes Regulations to Strengthen Consumers’ Privacy.

For employers, significant decisions can include decisions affecting hiring, compensation, promotion, discipline, or termination. That means an automated tool used in the employment lifecycle may trigger both:

• a civil-rights analysis under California employment-discrimination law; and
• a privacy analysis under California privacy and ADMT rules.

Those are related but not identical. A tool may raise discrimination concerns, privacy concerns, or both.

The privacy analysis may require employers to think about:

• pre-use notices;
• applicant and employee disclosures;
• access rights;
• opt-out rights where applicable;
• risk assessments;
• data minimization;
• retention periods;
• vendor data processing terms; and
• whether human review is meaningful or merely cosmetic.

Because the effective dates and obligations phase in, California employers should confirm the precise deadline and requirement before deploying or renewing an AI hiring tool.

California takeaway: treat AI in hiring as both an employment-law issue and a privacy issue.

Texas: The Intent-Based Approach

Texas took a very different path.

The Texas Responsible Artificial Intelligence Governance Act, commonly referred to as TRAIGA, was enacted through HB 149 and takes effect January 1, 2026. The Texas Legislature’s bill summary identifies the January 1, 2026 effective date, and the enrolled bill text identifies the statute as the Texas Responsible Artificial Intelligence Governance Act. Texas Legislature, HB 149 Bill Summary; HB 149 Enrolled Bill Text.

In the employment context, the key distinction is that TRAIGA’s discrimination prohibition is framed around intentional discrimination. Texas prohibits developing or deploying an AI system with the intent to unlawfully discriminate against a protected class under state or federal law. That is very different from California’s effects-based model.

Under California’s framework, a tool can create liability if it produces discriminatory effects, even without discriminatory intent. Under Texas’s TRAIGA framework, the state AI-law provision is narrower because intent is central.

That does not mean Texas employers can ignore disparate impact. It means the Texas AI statute itself is not the only law in the room.

What Texas Employers Do — and Do Not — Have to Do

Texas’s enacted law is more employer-friendly than many earlier AI-regulation proposals.

As a general matter, TRAIGA does not impose the same kind of broad employment-specific notice, public bias-audit, or formal impact-assessment obligations that employers may see in other jurisdictions. It also gives enforcement authority to the Texas Attorney General rather than creating a broad private right of action under the AI statute. The Texas Attorney General states that the AG has exclusive enforcement authority under the law. Texas Attorney General, Consumer AI Rights.

That said, Texas employers should not treat TRAIGA as permission to operate blindly.

The Attorney General can investigate. The statute includes cure procedures and civil penalties. The law also recognizes the relevance of established AI risk-management practices. In practical terms, employers that document reasonable governance, testing, oversight, and vendor diligence will be in a better position than employers that simply deploy tools and hope for the best.

Texas employers should still be able to answer:

• What AI or automated tools are used in hiring?
• Who selected the tool?
• What decisions does it influence?
• What data does it use?
• Has the tool been tested?
• Who reviews the results?
• Can a human override the recommendation?
• What happens if the tool appears to disadvantage a protected group?
• What does the vendor contract require?
• What records are retained?

Even if Texas law does not require the same compliance architecture as California, those questions remain essential risk-management questions.

TRAIGA Does Not Repeal Federal Law

This is the point Texas employers should not miss:

TRAIGA does not displace federal employment-discrimination law.

Title VII, the ADA, the ADEA, and other federal employment laws still apply. Federal law can reach hiring tools that disproportionately screen out protected groups, even where the employer did not intend to discriminate.

The EEOC has identified technology-related employment discrimination, including artificial intelligence and machine learning tools used in recruiting and hiring, as an enforcement priority. The EEOC also enforces federal employment-discrimination laws covering hiring, promotion, termination, compensation, and other employment decisions. EEOC Strategic Enforcement Plan Fiscal Years 2024–2028; EEOC Overview.

So a Texas employer should not stop at the question, “Did we intend to discriminate?”

The better question is:

Can we show that our hiring tool is job-related, consistent with business necessity, properly monitored, and not unlawfully excluding protected groups?

That is the safer question in every state.

The California–Texas Split

California and Texas now represent two very different models of AI employment regulation.

If you hire in both states — or employ remote workers in multiple states — do not design your process around the lowest standard.

Design it around the strictest applicable regime.

For many multistate employers, that means building the hiring process to satisfy California-style expectations: inventory the tools, understand the data, test for bias, preserve records, contract carefully with vendors, provide required notices where applicable, and keep humans meaningfully involved.

If your process can survive California scrutiny, it is usually in a stronger position elsewhere.

What Arizona Employers Should Know

Arizona has not enacted a California- or Texas-style statute specifically regulating AI in hiring as of this article’s publication date. Employers should still monitor Arizona legislative activity through official state sources. Arizona Secretary of State, Legislative Filings.

But Arizona employers are not in a rules-free zone. There are three reasons.

1. You May Be Hiring Into California or Texas

Many Arizona employers recruit remote workers, sales employees, engineers, executives, or operational staff across state lines. If the applicant is in California or Texas, the job is located there, or the employer’s hiring process reaches into those states, state-specific rules may become relevant. A Phoenix-based company can still create California risk if it screens California applicants through an automated tool.

2. Federal Law Applies Nationwide

Title VII, the ADA, the ADEA, and other federal employment laws apply regardless of whether Arizona has enacted a standalone AI hiring law. That means an Arizona employer using an automated selection tool should still care about:

• disparate impact;
• disability accommodation;
• age discrimination;
• race, sex, religion, color, and national-origin discrimination;
• validation of selection procedures;
• job-relatedness;
• business necessity;
• recordkeeping; and
• EEOC scrutiny.

3. Vendors Do Not Configure Risk by State Unless You Require It

Many AI hiring tools are sold nationally. The same platform may be used for California, Texas, Arizona, and remote roles. If the vendor has not configured the tool by jurisdiction, the employer may be applying the same risky process everywhere. Arizona employers should therefore build a baseline AI hiring program even if Arizona has not enacted a dedicated statute.

What Is Still Coming?

AI employment regulation is moving quickly.

California is not done. In the 2025–2026 legislative session, lawmakers considered additional bills targeting automated decision systems. SB 7 addressed employment automated decision systems but did not become law. California SB 7 Bill Page. AB 1018, the Automated Decision Systems bill, was ordered to the inactive file in September 2025. California AB 1018 Status. In 2026, SB 947 was introduced to address employment automated decision systems. California SB 947 Bill Page.

The details may change, but the direction is clear:

• more disclosure;
• more documentation;
• more human oversight;
• more attention to bias testing;
• more vendor accountability;
• more recordkeeping; and
• more scrutiny of automated tools used in consequential workplace decisions.

Employers should not wait for every bill to become final before acting. A basic governance program is already prudent under existing California, Texas, and federal law.

A Practical Checklist for Employers Using AI in Hiring

If your company uses AI or automated tools in hiring, start here.

1. Inventory Every Tool

List every system that touches:

• recruiting;
• resume screening;
• applicant ranking;
• interview scheduling;
• video interviews;
• assessments;
• background-screening workflows;
• candidate recommendations;
• promotion decisions;
• compensation decisions;
• discipline; or
• termination.

Do not rely only on what HR calls “AI.” Ask what the system actually does.

2. Map the Decision Points

For each tool, identify:

• when it is used;
• who uses it;
• what decision it influences;
• whether it excludes candidates;
• whether it ranks candidates;
• whether it generates a score;
• whether a human reviews the output;
• whether the human can override the result; and
• whether the override is documented.

A tool that merely organizes applications is different from a tool that decides who gets interviewed.

3. Ask Vendors Better Questions

Do not accept marketing claims. Ask for documentation. Important vendor questions include:

• What data does the tool collect?
• What data does the tool infer?
• What data does the tool use to score, rank, or recommend?
• Has the tool been validated for the job at issue?
• Has the tool been tested for adverse impact?
• What groups were tested?
• When was the testing performed?
• Can the vendor provide testing summaries?
• Can the vendor support litigation holds and agency inquiries?
• Can the employer access scores, rankings, criteria, and historical outputs?
• Does the vendor use applicant data to train models?
• Can applicant data be deleted or segregated?
• Does the vendor contract include audit rights?
• Does the contract include indemnity for tool-related legal failures?

If the vendor cannot answer, that is an answer.

4. Run Anti-Bias Testing

Testing should be more than a one-time checkbox. Employers should test:

• before deployment;
• after significant tool changes;
• after changing job criteria;
• after expanding into new jurisdictions;
• after a complaint;
• after unusual selection-rate patterns; and
• periodically as part of normal governance.

Testing should be paired with action. If the tool creates adverse impact, investigate and fix the issue.

5. Keep Humans Meaningfully in the Loop

A human in the loop is not helpful if the human simply rubber-stamps the machine. Meaningful human review means the reviewer:

• understands what the tool does;
• knows the limits of the tool;
• can question the output;
• can override the recommendation;
• documents the reason for the decision; and
• is trained not to treat the AI score as automatically correct.

6. Update Notices and Privacy Documents

For California applicants and employees, employers should review whether privacy notices, applicant notices, internal policies, and ADMT disclosures need to be updated. Do not bury AI disclosures in vague privacy language. If a tool materially affects hiring or other employment decisions, applicants and employees may need clear information about how the tool is used and what rights they have.

7. Fix Recordkeeping Before There Is a Dispute

Employers should confirm they can preserve:

• the version of the tool used;
• job criteria;
• applicant inputs;
• scores;
• rankings;
• recommendations;
• exclusions;
• human-review notes;
• testing results;
• vendor communications;
• notices;
• consent or acknowledgment records where applicable; and
• audit results.

A recordkeeping policy that depends on a vendor’s default settings is not enough.

8. Write an AI-in-Hiring Policy

A good policy should identify:

• approved tools;
• prohibited tools;
• who may approve new tools;
• what review is required before deployment;
• when legal review is required;
• when bias testing is required;
• how human review works;
• what records must be kept;
• how applicants can raise concerns;
• how accommodations are handled; and
• who owns ongoing monitoring.

The policy should be short enough for HR and hiring managers to use, but specific enough to matter.

9. Train HR and Hiring Managers

The best policy will fail if managers do not understand it. Training should cover:

• what tools the company uses;
• what the tools are allowed to do;
• what the tools are not allowed to do;
• how to document human review;
• how to spot potential bias;
• how to handle accommodation requests;
• how to escalate complaints; and
• when to involve legal.

10. Review the Program Regularly

AI hiring compliance is not a one-and-done project. Tools change. Vendors update models. Laws change. Job requirements change. Applicant pools change. Enforcement priorities change. Employers should review their AI hiring program at least annually and whenever they adopt or materially change a tool.

Frequently Asked Questions

Final Takeaway

AI hiring tools are not illegal. Unexamined AI hiring tools are the problem.

California expects employers to pay close attention to discriminatory effects, documentation, vendor use, recordkeeping, and privacy obligations. Texas takes a narrower, intent-based approach under its AI statute, but Texas employers still need to account for federal law and responsible AI governance. Arizona employers may not have a standalone AI hiring statute yet, but they are still subject to federal law and multistate hiring risk.

For employers, the safest approach is practical and disciplined:

• know what tools you use;
• understand what they do;
• test them;
• document the process;
• update notices where required;
• preserve the records;
• train the humans; and
• tighten the vendor contracts.

AI may help you hire faster. Compliance helps make sure you can defend how you hired.

Sources

This article draws on the following primary and agency sources:

• California Civil Rights Department — Civil Rights Council Rulemaking Actions (Employment Regulations Regarding Automated Decision Systems, effective October 1, 2025)
• California Privacy Protection Agency — ADMT regulations; “California Finalizes Regulations to Strengthen Consumers’ Privacy”
• Texas Legislature — HB 149 Bill Summary and Enrolled Bill Text (Texas Responsible Artificial Intelligence Governance Act, effective January 1, 2026)
• Texas Attorney General — Consumer AI Rights
• U.S. EEOC — Strategic Enforcement Plan, Fiscal Years 2024–2028; EEOC Overview
• California Legislature — SB 7, AB 1018, and SB 947 bill pages and status
• Arizona Secretary of State — Legislative Filings