You're hiring for summer staff. You've got 500 applications. Your hiring team can't read them all. So you deploy an AI resume screener. It's fast, it's efficient, it sorts resumes in seconds. You run it and get 50 candidates who "match" your criteria.

You never audit the tool for bias. You never disclosed to candidates that an algorithm was making decisions about their applications. You never checked whether the algorithm was filtering out protected classes at a disproportionate rate.

Three years later, a pattern emerges. Your workforce is 95 percent white and male. Your screener filtered out 73 percent of female candidates and 68 percent of candidates over 55. The EEOC files a lawsuit. Your tool, trained on historical hiring data, learned to replicate the discrimination that already existed in your workforce.

Welcome to the brave new world of AI hiring liability, where automation doesn't eliminate bias. It scales it.

Before we go deeper, let's establish what we're talking about. An "automated employment decision tool" (AEDT) is any artificial intelligence, algorithm, machine learning model, or statistical analysis used by an employer to automate, inform, or influence an employment decision. That includes resume screeners, interview scheduling systems, video assessment tools, chatbots, background check analysis, and predictive hiring software.

The legal problem is that these tools can violate federal civil rights laws (Title VII, ADA, ADEA) and state and local laws like New York's Local Law 144 without you knowing it. Because the tool is automated, the discrimination can be massive and systematic before anyone notices.

Here's a risk scorecard to assess your exposure. Rate each category on a scale of 1 (low risk) to 5 (high risk):

1. Resume Screening and Initial Filtering (Highest Risk)

Does your company use AI to screen resumes, filter by keywords, or automatically reject applications before human review? Do you know what criteria the tool uses? Have you conducted a bias audit? Do candidates know an algorithm is screening their applications?

Risk factors: This is a tool trained on historical data that reflects past discrimination. Algorithms can learn to filter based on proxies for protected class status (employment gaps disproportionately affect women and disabled workers; certain schools or neighborhoods correlate with race). No human review layer. No candidate disclosure.

2. Interview Scheduling and Coordination

Are you using AI to schedule interviews, screen for availability, or rank candidates for interview order? Does the tool consider factors that correlate with protected class status (time zone preferences, availability flexibility, response speed)?

Risk factors: An algorithm that prioritizes immediate responders might systematically disadvantage parents or disabled workers who need scheduling flexibility. A tool trained on past hiring might learn to schedule women or older workers at inconvenient times, creating barriers.

3. Assessment and Interview Tools

Are you using AI-powered assessments, video interview platforms that use facial recognition or emotion detection, coding evaluators, or personality tests that are scored by algorithm? Do you understand what the algorithm is measuring and whether it's job-related?

Risk factors: Video interview platforms that score facial expressions, eye contact, or speech patterns can discriminate against neurodiverse candidates, candidates with disabilities, non-native speakers, and candidates from different cultural backgrounds. Personality tests scored by AI can have disparate impact if not validated for job relevance.

4. Background Check Analysis

Are you using algorithms to analyze background check results, reject candidates based on certain criminal records, or flag candidates for further review? Does the algorithm consider the relationship between the record and the job?

Risk factors: An algorithm that automatically rejects anyone with a criminal history has disparate impact on Black and Latino candidates (who are overrepresented in the criminal justice system) and may violate the EEOC's ban-the-box guidance even if your policy is facially neutral.

5. Chatbots and Preliminary Screening

Are you using chatbots to conduct preliminary conversations with candidates, screen for basic qualifications, or assess culture fit? Does the bot ask questions about disability status, age, family, or other protected characteristics?

Risk factors: A well-meaning chatbot can ask "Do you have any disabilities that would prevent you from working?" in violation of the ADA. A bot trained on past hiring data might ask different questions of different candidates based on their perceived background, creating disparate treatment.

6. Predictive Hiring and "Culture Fit" Algorithms

Are you using tools that predict which candidates will "succeed" at your company, score candidates on "culture fit," or use algorithms to identify candidates who will be long-term employees? Do you understand what data the tool uses to make these predictions?

Risk factors: Predictive tools trained on employee retention data learn to hire people who are similar to your current employees. If your current workforce lacks diversity, the tool will optimize for homogeneity. A "culture fit" algorithm is discrimination waiting to happen. Culture fit is not a lawful job qualification.

In 2023, New York City enacted Local Law 144. It became effective on January 1, 2024. If you're operating in New York City and using an AEDT for hiring decisions, it applies to you.

Here's what Local Law 144 requires:

1. Bias Audit

Before you use an AEDT to hire in New York City, you must conduct a bias audit. The audit must assess whether the tool has disparate impact on people of color, women, or other protected groups. The audit must be conducted by someone who isn't affiliated with the tool's developer. The audit must be recent (generally within one year of deployment).

Translation: You can't just buy an off-the-shelf hiring tool and deploy it. You have to prove it's not discriminatory before you use it.

2. Notice to Candidates

You must notify candidates, in writing, that you're using an AEDT to make or influence hiring decisions. You must tell them what kind of data the tool uses. You must explain how they can request a manual review if they want a human to look at their application instead of relying on the algorithm.

Translation: No more secret algorithms. Transparency is required.

3. Manual Review Option

If a candidate requests it, you must have a human being conduct a manual review of their application or interview. You can't force candidates to accept algorithmic decisions.

Translation: The algorithm informs the decision, but humans make the final call (at least on request).

4. Record-Keeping

You must keep records of your bias audits and retain them for the life of the tool's use, plus two years after you stop using it.

Translation: Be prepared to explain yourself to investigators.

5. Approval by City

The Department of Consumer and Worker Protection (DCWP) and the Commissioner of Human Rights must approve the bias audit and can issue guidance on how tools should be tested. If they determine a tool has unlawful disparate impact, they can order you to stop using it.

Translation: New York City is the enforcement arm here, and they're watching.

Violating Local Law 144 results in penalties up to $1,000 per day that the violation occurred, plus attorney fees. It's a strict liability standard, which means even innocent mistakes can result in penalties.

But Local Law 144 isn't just a New York problem. The EEOC has also weighed in.

In 2023, the EEOC issued guidance on how AI hiring tools are evaluated under Title VII of the Civil Rights Act. The guidance confirms what employment law folks have known for a while: an algorithm that produces discriminatory outcomes violates Title VII, even if no human intended discrimination and even if the discrimination wasn't explicit.

Here's what the EEOC said:

1. Automated Tools Can Violate Title VII

If a tool has disparate impact on a protected class (race, color, religion, sex, national origin, disability, age), it violates Title VII, even if the tool itself doesn't mention protected class status.

2. Training Data Matters

If a tool is trained on historical data that reflects past discrimination, it will learn to discriminate. An algorithm trained on your past hiring (where you hired mostly men for engineering roles, mostly women for administrative roles) will perpetuate that pattern.

3. Validation Is Required

An employer should validate that an automated tool actually measures what it purports to measure and that it's job-related. Just because a tool correlates with past success doesn't mean it's a valid measure. A resume screener that filters out employment gaps looks objective, but if it has disparate impact on women (who are more likely to have gaps due to caregiving responsibilities), it's discriminatory.

4. Ongoing Monitoring Is Required

Employers using automated tools must monitor them for disparate impact. An algorithm that worked fine for three years might develop bias as it's applied to new populations or as the job market changes. You need to check.

The EEOC is willing to sue based on disparate impact from AI hiring tools. A pattern of AI-driven discrimination can be the basis for major damages awards and consent decrees.

Let's talk about what a real bias audit looks like, because this is where a lot of employers get it wrong.

A bias audit is not a check-the-box exercise. It's a statistical analysis of whether the tool produces different outcomes for different groups.

Here's the basic process:

Step 1: Define "Disparate Impact"

Regulatory agencies use the "four-fifths rule" as a rough benchmark. If a protected group is selected (hired, advanced, invited to interview) at less than 80 percent of the rate of the most-favored group, there's potential disparate impact and the tool needs explanation.

For example: If your resume screener selects 40 percent of male applicants but only 20 percent of female applicants, that's 50 percent of the male selection rate. That's obvious disparate impact.

Step 2: Collect Data on Tool Decisions

You need to know how the tool treated different groups. Did it accept or reject more applications from women than men? More from older workers than younger workers? More from white candidates than candidates of color?

This requires tracking protected class data (race, sex, age, disability status) in your applicant flow, which employers often don't do. You need to go back and collect it.

Step 3: Run the Statistics

The audit compares selection rates by protected group and calculates whether disparities are statistically significant (not just random chance).

Step 4: Assess the Tool's Predictive Validity

Does the tool actually measure job performance? A resume screener that filters by certain schools might have disparate impact and also be invalid (going to a top-tier school doesn't predict job performance for many roles). An assessment tool that screens based on personality traits might be invalid for the job. Invalid tools are indefensible.

Step 5: Recommend Adjustments or Discontinuation

A real audit concludes with recommendations. Maybe the tool can be retrained on better data. Maybe a human review layer can be added. Maybe the tool should be abandoned entirely.

The cost of a real bias audit: $5,000 to $30,000+, depending on the tool's complexity and your applicant volume. The cost of deploying a discriminatory tool: unlimited litigation liability.

Let's make this concrete. Here's how a seemingly innocent hiring tool creates liability:

Resume Screeners and Employment Gaps

You deploy a resume screener to quickly filter applications. The screener is trained on data from your past hiring, looking for resumes that correlate with people who were hired. The tool learns to favor:

Now, here's the problem: Women are statistically more likely to have employment gaps due to caregiving responsibilities. The screener learns to filter out resumes with gaps. It disproportionately rejects female candidates, some of whom are highly qualified. That's disparate impact, and it's actionable discrimination.

The fact that the tool doesn't mention "gender" is irrelevant. The outcome is discriminatory.

Video Interview Tools and Facial Recognition

You deploy a video interview platform that uses AI to analyze candidates' responses. The tool scores candidates on "engagement," "confidence," and "eye contact." It's supposed to standardize interview scoring.

Here's the problem: Eye contact is culturally dependent. In some cultures, sustained eye contact with an authority figure is disrespectful. The tool penalizes candidates from those backgrounds. Facial expression analysis is notoriously biased against people with disabilities (autism spectrum, Parkinson's, Bell's palsy) who have atypical expressions. The tool discriminates against neurodiverse candidates.

Disparate impact on disability status. That's an ADA violation.

Chatbots and Illegal Screening Questions

You deploy a chatbot to do preliminary screening. The bot asks a series of standard questions to all applicants. But the bot is trained to ask follow-up questions based on candidate responses. One candidate mentions they're returning to work after a few years away. The bot asks, "Why were you out of the workforce?" Another candidate mentions they're managing multiple jobs. The bot asks, "Do you have reliable childcare?"

Different candidates are asked different questions. The questions reveal family status and caregiving responsibilities, which are protected under Title VII and various state laws. Some candidates are screened out based on this information.

Disparate treatment (different questions for different candidates) based on protected status. That's discrimination.

Personality Assessments and "Culture Fit"

You deploy a personality assessment tool to identify candidates who match your company culture. The tool scores candidates on traits like "openness," "conscientiousness," and "agreeableness." Candidates who score high are flagged as strong fits. Candidates who score low are rejected.

Here's the problem: Personality is measured differently across cultures and demographics. Traits that seem like positive "agreeableness" in some populations might look like lack of assertiveness in others. An introversion preference is not a lawful job qualification, but if your culture happens to be extroverted and your hiring has been predominantly extroverted, the tool learns to prefer extroverts.

If extroversion correlates with certain demographic groups and introversion with others (research suggests some evidence for this), the tool creates disparate impact based on protected status, not job qualification.

You intended to find culture fit. You created a machine that replicates your existing biases.

^{[Note: This is a hypothetical scenario based on common compliance failures.]}

JobSpeed Staffing is a mid-size staffing agency in New York with 50 employees. They place temporary workers at client companies across hospitality, retail, and administrative sectors. They place more than 300 workers annually.

Three years ago, JobSpeed implemented an AI resume screener to handle the volume. The tool was built on machine learning: feed it historical data about which candidates they'd hired, and it learns to identify "high-quality" candidates. JobSpeed fed the tool three years of hiring data and let it go.

The tool was fast. Applications that used to take two days to review now took two hours. JobSpeed could screen more applicants and fill positions faster. It seemed like a clear win.

No one conducted a bias audit. No one disclosed the tool to candidates. No one monitored the outcomes.

Then a job seeker files a discrimination complaint with the New York City Commission on Human Rights (CCHR). She's a woman, 58 years old, with a 10-year employment gap due to caring for a parent. Her resume is strong: prior experience in the same field, relevant skills, available immediately. But she keeps getting rejected after applying to JobSpeed placements.

She applies 12 times over three months and gets rejected 12 times. Her equal-aged male colleague applies to the same positions and gets 8 interviews out of 12 applications.

She files a complaint. CCHR opens an investigation and asks JobSpeed: "What AI tools do you use to screen applications?"

JobSpeed provides documentation. The tool was trained on three years of hiring data. CCHR investigates the outcomes.

The analysis reveals:

Employment gaps are disproportionately common among women (due to caregiving) and older workers (due to health issues, caregiving, industry disruption). The tool learned to penalize them.

The selection gaps are far below the four-fifths threshold. There's clear disparate impact.

CCHR looks further and finds that JobSpeed:

CCHR issues a violations finding and orders JobSpeed to:

Meanwhile, the EEOC independently investigates, finding similar patterns across JobSpeed's placements. The agency determines there's evidence of disparate impact discrimination in violation of Title VII. They file suit for systemic discrimination.

JobSpeed faces:

The tool that was supposed to make hiring more efficient created a machine that systematically discriminated against women and older workers for three years, affecting hundreds of job placements.

^{[Note: This is a hypothetical scenario based on compliance best practices.]}

Beacon Hospitality is a regional hotel group operating four properties in the Northeast, employing about 200 people total. The hotels are in a tight labor market, and turnover is a persistent challenge. To improve hiring efficiency, they decided to implement AI scheduling tools to reduce time-to-hire and improve retention predictions.

Before deploying any tool, Beacon did the right thing:

1. Vendor Assessment

They requested that AI vendors provide recent bias audits completed by independent third parties. They reviewed the audit methodology and results. They asked vendors hard questions about disparate impact findings. They rejected vendors who wouldn't provide transparent audit results.

2. Pre-Deployment Audit

Before going live with the selected scheduling tool, Beacon commissioned their own bias audit from an independent firm. The audit analyzed the tool's performance on test datasets representing diverse candidate pools (race, sex, age, disability). The audit found no statistically significant disparate impact.

Cost: $8,000 for the audit.

3. Candidate Notice and Disclosure

Beacon created a standard notice explaining that they use AI tools to optimize interview scheduling. The notice explained what data the tool uses and informed candidates that they could request a manual scheduling option if they preferred human handling.

The notice was provided to all candidates in writing, before the tool was used.

4. Human Review Layer

Beacon implemented a policy requiring human review of any candidate who requested it. More importantly, they added a human review layer for edge cases: candidates with disability-related accommodation requests, candidates with scheduling constraints they flagged as important, and any candidate who was rejected by the tool but had borderline metrics.

The human review layer caught cases where the tool was technically correct but missed context the human would catch.

5. Ongoing Monitoring

Beacon implemented quarterly audits of the tool's performance. Each quarter, they pulled applicant flow data by race, sex, age, and disability status and compared selection rates. They monitored for unexpected changes in applicant demographics.

When they discovered that the tool was scheduling older candidates for less convenient times (early mornings and late evenings, when fewer people want to interview), they adjusted the algorithm.

6. Documentation and Record-Keeping

Beacon kept meticulous records of all audits, monitoring, and adjustments. They documented their decision-making process. If an investigator ever asked, they could show exactly what they did and why.

7. Training

Beacon trained hiring managers on how to use the tool responsibly. They explained that the tool informs decisions but doesn't make them. They emphasized the importance of human judgment in edge cases.

Three years in, Beacon's AI hiring implementation is compliant with Local Law 144 and EEOC expectations. The tool improved their time-to-hire by 20 percent and helped reduce turnover. Because they monitored for bias and adjusted for edge cases, they didn't create a discrimination machine.

Beacon's ongoing costs for the tool (including audits and monitoring): roughly $15,000 per year.

The risk they avoided: hundreds of thousands in discrimination liability, civil penalties, reputational damage, and litigation.

One of the most-missed compliance requirements under Local Law 144 is the notice requirement. Many employers don't realize they have to tell candidates that an algorithm is reviewing their applications.

If you're using an AEDT in New York City, you must provide:

Clear Written Notice

The notice must explain that you're using an automated decision tool. It can't be buried in fine print. It should be clear, prominent, and easy to understand.

What Data the Tool Uses

You must tell candidates what categories of data the tool considers. For example: "We use resume keywords, years of experience, and educational background to screen applications."

How to Request Manual Review

You must explain how candidates can request that a human being review their application or interview instead of relying on the algorithm. The process should be simple and should not disadvantage the candidate.

Impact on Application Status

You should explain that the tool informs decisions but doesn't make final decisions alone (if that's true, and it should be).

The notice should be provided to all candidates in writing. If you're applying to a job online, the notice should appear before they submit their application. If you're accepting applications by email, it should be in your initial communication.

Failure to provide notice is a violation of Local Law 144 regardless of whether the tool is actually biased. Transparency is required.

Here's a practical checklist to assess your own exposure and begin addressing it:

1. Inventory Your Tools

Do you use any AI, algorithms, or automated systems in your hiring pipeline? Resume screeners, scheduling tools, assessment platforms, video interview tools, chatbots, background check analyzers, or predictive hiring software all count.

2. Identify Your Legal Obligations

Are you operating in New York City? If yes, Local Law 144 applies. Does your hiring operate under federal contracts? If yes, you may have additional OFCCP obligations. In any case, Title VII applies to companies with 15+ employees. The ADEA applies to companies with 20+ employees.

3. Request Vendor Audits

For each tool you use, request a recent bias audit from the vendor. Ask for the complete findings, not just a summary. If the vendor won't provide an audit or the audit is more than a year old, that's a problem.

4. Conduct Your Own Audit

Don't rely solely on vendor audits. Commission an independent third-party bias audit of your tools, especially if you've been using them for years. The audit should analyze disparate impact by race, sex, age, and disability status.

5. Implement Transparency

Create notices for all candidates explaining your use of AI tools. Explain how candidates can request manual review. Make the request process simple and non-burdensome.

6. Add Human Review

Don't let the algorithm make final decisions. Have humans verify edge cases, candidates with accommodation requests, and any applicants flagged by candidates themselves.

7. Monitor Continuously

Set up quarterly or annual reviews of your tool's outcomes. Pull applicant flow data by protected class. Compare selection rates. If you see disparities, investigate.

8. Document Everything

Keep records of your audits, monitoring, decisions, and adjustments. Documentation is your defense if you ever face investigation.

9. Train Your Team

Make sure hiring managers understand how to use the tools responsibly. Emphasize that the tool informs, doesn't dictate.

10. Budget for Compliance

A comprehensive bias audit costs $5,000-$30,000. Ongoing monitoring and annual audits cost $2,000-$10,000 per year. Add disclosure, human review processes, and training. Budget for compliance. It's cheaper than litigation.

AI hiring tools are proliferating. The technology is getting more sophisticated. And most employers deploying these tools have no idea what legal exposure they're creating.

The tools themselves aren't inherently illegal. Automation isn't inherently discriminatory. But automation without oversight, without auditing, without transparency, and without human judgment creates systemic discrimination that's hard to defend and easy to miss.

The legal landscape is shifting. New York has taken the lead with Local Law 144, but other jurisdictions are following. The EEOC is signaling that it will aggressively pursue disparate impact cases involving AI tools. The FTC is investigating AI hiring tools for potential unfair and deceptive practices. If you're not thinking about your AI hiring liability now, you will be soon.

The good news: Compliance is achievable. It requires investment in auditing, transparency, and monitoring, but it's doable. The companies getting this right are the ones that treat AI as a tool that informs human judgment, not replaces it. They audit their tools. They tell candidates about them. They monitor for bias. And they maintain human oversight at critical decision points.

If you're using AI in your hiring pipeline, audit it now. Don't wait for an investigation. Don't assume it's compliant just because it's from a reputable vendor. Do the work. It will save you a lot of pain down the road.

For more on hiring compliance and discrimination issues, see Signs of the Times.

Keep fighting the good fight.

^{LEGAL DISCLAIMER}

^{This article is informational only and does not constitute legal advice. Employment law is complex and fact-specific. The examples, case studies, and analyses in this article are hypothetical and are not a substitute for advice from an attorney. Every employer's situation is different, and compliance with AI hiring tools, federal civil rights laws, and local regulations like NYC Local Law 144 requires careful review of your specific facts and applicable law.}

^{This article contains advertising material. The information provided is intended for general informational purposes and is not a substitute for legal counsel.}

^{Copyright 2026 Jacobs & Associates LLC. All rights reserved.}

^{Jacobs & Associates LLC is a solo employment law practice in Jersey City, New Jersey, representing employees and employers in workplace disputes, compliance matters, and litigation.}