Job-market ghosting traditionally means one side suddenly stops communicating.
An applicant:
- Applies for a position.
- Completes an assessment.
- Uploads a résumé.
- Records a video interview.
- Answers chatbot questions.
Then:
Nothing.
No recruiter calls.
No explanation arrives.
The candidate may simply receive an automated rejection—or hear nothing at all.
AI hasn’t created candidate ghosting, but large-scale automated recruiting can make it easier for businesses to process enormous applicant pools with very little human interaction.
That creates an important risk-management problem.
The more hiring decisions you automate, the more important it becomes to understand:
how those decisions are actually being made.
What Is EPLI?
Employment Practices Liability Insurance, commonly called EPLI, is designed to address certain employment-related claims.
Depending on the policy, allegations can include:
- Discrimination
- Harassment
- Wrongful termination
- Retaliation
- Failure to promote
- Certain hiring-related claims
- Other specified employment practices
Policy terms differ considerably.
An EPLI policy doesn’t mean:
“Anything involving an employee is covered.”
And AI-related claims can create additional questions involving definitions, exclusions, regulatory proceedings and third-party technology.
The New Hiring Department May Be an Algorithm
Consider a company receiving:
8,000 applications
for:
100 positions.
Human recruiters cannot realistically conduct detailed initial reviews of every applicant.
So the company deploys an AI-assisted hiring platform.
The system:
8,000 applicants
↓
2,000 candidates pass initial screening
↓
500 receive automated assessments
↓
200 receive interviews
↓
100 hired
Efficient?
Absolutely.
Risk-free?
No.
The key question becomes:
What caused the algorithm to eliminate the other 7,900 people?
AI Can Screen Résumés
Employers increasingly use automated tools for functions such as:
- Résumé screening
- Candidate ranking
- Skills matching
- Assessment scoring
- Interview analysis
- Recruiting chatbots
The EEOC specifically identifies résumé keyword screening and recorded-video interview evaluation among examples of AI use that workers may encounter during hiring.
Automation itself isn’t automatically discriminatory.
The risk is that:
inputs + training data + model design + employer configuration
can potentially produce unlawful outcomes.
Example: The Historical Hiring Problem
Imagine a company trains a hiring model using data from its:
highest-performing employees over the past ten years.
That sounds reasonable.
But suppose historical hiring practices resulted in one demographic group being heavily overrepresented.
The model identifies patterns associated with those employees.
It may then favor candidates displaying similar patterns.
The employer never programs:
“Reject Group X.”
But the system may still produce disparities.
This illustrates why:
intent
and
outcome
aren’t necessarily the same thing.
Existing Discrimination Laws Still Apply
AI does not operate outside employment law.
The EEOC states that federal employment-discrimination protections continue to apply when employers use AI.
Protected characteristics under federal law can include:
- Race
- Color
- Religion
- Sex
- National origin
- Age 40 or older
- Disability
- Genetic information
depending on the particular statute.
An employer generally cannot defend an unlawful hiring practice merely by saying:
“The algorithm selected the candidates.”
Disparate Impact Is an Important Risk
Employment discrimination isn’t limited to explicitly telling a system:
“Don’t hire women.”
A seemingly neutral selection procedure can also create legal problems when it disproportionately excludes members of a protected group and cannot be appropriately justified under applicable law.
The EEOC explains that employment tests and selection procedures can violate federal anti-discrimination laws when they disproportionately exclude people based on a protected characteristic unless the employer can justify the procedure under applicable legal standards.
AI screening tools therefore deserve the same scrutiny as other employee-selection procedures.
Disability Can Create a Different AI Problem
Consider a candidate with a disability.
The employer uses an automated video assessment.
The system evaluates behavioral characteristics that are supposedly associated with successful employees.
But the candidate’s disability affects how they interact with the software.
The algorithm gives them a poor score.
They are rejected.
The candidate may have been perfectly capable of performing the actual job.
The EEOC has specifically warned that algorithmic hiring tools may unlawfully screen out people with disabilities who could perform a job with or without reasonable accommodation.
Reasonable Accommodation Still Matters
An employer using AI should have a process for candidates who need accommodation.
The EEOC identifies reasonable accommodation as one of the key concerns when employers use algorithmic decision-making tools.
For example, depending on the circumstances, a candidate might need:
- Alternative assessment format
- Additional time
- Accessible technology
- Human-assisted process
- Alternative method of demonstrating ability
Businesses should ensure that applicants know how to request appropriate accommodations.
“Our Vendor Built It” Is Not a Complete Defense
This may be one of the most important lessons for businesses.
Imagine an employer buys an AI recruiting platform.
The vendor says:
“Our algorithm is unbiased.”
The employer activates it.
Six months later, applicants allege discrimination.
Can the employer simply say:
“Talk to the software vendor”?
Not necessarily.
New York City guidance, for example, states that employers, employment agencies and their agents can face liability under city human-rights law for discrimination resulting from technology or AI.
Vendor selection therefore becomes part of employment-risk management.
Ask Your AI Vendor Difficult Questions
Before deploying a hiring tool, businesses should understand:
- What employment decisions does the system influence?
- What data does it collect?
- How are candidates scored?
- What variables affect ranking?
- Has the tool been tested for discriminatory outcomes?
- How frequently is it tested?
- What accommodation options exist?
- Can a human override the recommendation?
- How are overrides documented?
- How long is candidate data retained?
- Does the vendor use customer data for model training?
- What happens when the model changes?
- Who investigates suspected errors?
- What contractual protections does the vendor provide?
- Will the vendor cooperate during litigation?
If the vendor cannot clearly answer basic questions about a system influencing hiring decisions, that is itself a risk signal.
New York City Already Regulates Automated Hiring Tools
New York City provides one of the clearest examples of specific AI hiring regulation.
Local Law 144 restricts employers and employment agencies from using covered Automated Employment Decision Tools (AEDTs) unless certain requirements are satisfied.
Among them:
Bias audit
The tool must have undergone the required bias audit within the applicable period.
Public disclosure
Information concerning the audit must be made publicly available.
Notice
Covered candidates and employees must receive required notice.
Enforcement began in July 2023.
For employers operating in multiple jurisdictions, the lesson is important:
One nationwide AI hiring process may encounter different local requirements.
A Bias Audit Isn’t a Magic Shield
Suppose your tool passes a required audit.
Does that mean:
“Zero liability forever”?
No.
An audit is a compliance mechanism—not immunity from every employment claim.
Businesses still need to consider:
- Federal anti-discrimination law
- State requirements
- Local requirements
- Disability accommodation
- Actual hiring outcomes
- Changes to the algorithm
- Changes to candidate populations
AI governance needs to be ongoing.
“Ghosting” Can Hide Patterns
Imagine 10,000 people apply.
The system rejects:
8,500 automatically.
Nobody regularly examines those rejected candidates.
That’s risky.
The rejected population may contain the most important information about whether your hiring process creates unintended disparities.
Employers should consider monitoring:
Who applies?
Who passes each stage?
Who is rejected?
Who receives interviews?
Who receives offers?
Who accepts?
The goal isn’t simply measuring:
time-to-hire.
It is understanding the hiring funnel.
Human Review Helps—but Isn’t Magic Either
Some businesses respond:
“Don’t worry. A human makes the final decision.”
That’s useful, but it doesn’t automatically eliminate algorithmic risk.
Suppose:
AI reviews 10,000 candidates.
It eliminates:
9,500.
Human recruiters only review:
500.
Humans technically make the final hiring decisions.
But the AI determined who was allowed to reach the humans.
The meaningful employment decision may therefore occur much earlier in the process.
Automation Bias Is Another Problem
Humans can place too much confidence in automated recommendations.
Recruiter sees:
Candidate A — 94% match
Candidate B — 67% match
The recruiter may assume:
94% = objectively better.
But what exactly does that number represent?
A score can appear scientific while hiding assumptions about:
- Training data
- Job requirements
- Candidate history
- Weighting
- Statistical correlations
Human oversight only works when humans are genuinely empowered to question the machine.
Train Recruiters to Challenge AI
Recruiters should understand:
AI recommendation ≠ command.
Training should explain:
- What the tool does
- What it doesn’t do
- What data it uses
- Known limitations
- When human review is necessary
- How accommodations work
- How to escalate unusual outcomes
- How overrides are documented
Otherwise “human oversight” may exist only on paper.
Candidate Notice Matters
Transparency is increasingly important.
New York City’s rules require notice in covered situations, including informing candidates or employees that an AEDT will be used.
Even where a particular notice law doesn’t apply, businesses should consider whether applicants clearly understand:
when automation is materially involved in evaluating them.
Transparency can also help candidates identify when they need an accommodation.
What Does This Have to Do With EPLI?
Everything.
An unsuccessful applicant might allege:
“Your AI hiring system discriminated against me.”
That can potentially become an employment-practices claim.
The employer then needs to determine:
Does our EPLI policy respond?
Don’t wait for litigation to find out.
Does EPLI Cover Applicants?
This is a crucial policy question.
Some EPLI policies may define covered claimants broadly enough to include:
job applicants.
Others may contain different definitions, limitations or endorsements.
Ask your broker:
“Does our EPLI policy cover discrimination claims brought by applicants who were never employed by us?”
For AI-heavy hiring organizations, this should be specifically confirmed.
Regulatory Investigations May Be Different
Suppose an individual files a discrimination lawsuit.
Your EPLI may potentially respond according to policy terms.
But what happens when a:
regulator
investigates the hiring system?
Coverage for:
- Government investigations
- Administrative proceedings
- Fines
- Penalties
- Compliance costs
can differ considerably.
Certain fines or penalties may also be uninsurable under applicable law.
Businesses should understand these distinctions before a regulator arrives.
AI Can Create Class-Wide Exposure
Traditional employment disputes may involve:
one employee.
Algorithmic systems can affect:
thousands of candidates simultaneously.
Imagine one screening rule is flawed.
It processes:
50,000 applicants.
Even a small statistical problem can potentially affect a large population.
That creates a defining characteristic of AI risk:
automation scales mistakes.
A human recruiter can make one poor decision.
An algorithm can potentially repeat a problematic decision:
every few seconds.
This Can Affect EPLI Limits
Suppose your business has:
$1 million EPLI limit.
That may seem substantial.
But imagine allegations involving:
- Thousands of applicants
- Class litigation
- Regulatory investigation
- Expert witnesses
- Algorithmic analysis
- Discovery
- Defense costs
The economics can look very different.
Businesses using AI extensively should discuss whether existing EPLI limits remain appropriate.
Defense Costs Matter
Read how your policy treats legal defense expenses.
Some EPLI policies may have defense costs:
inside the limit.
For example:
$1,000,000 limit
minus:
$300,000 defense costs
leaves:
$700,000
for other covered amounts.
Policy structures vary.
For complex AI-related litigation, defense-cost treatment can become particularly important.
Review Your Deductible or Retention
EPLI policies commonly involve a deductible or self-insured retention.
For example:
EPLI limit: $2 million
Retention: $50,000
The business may therefore bear the first:
$50,000
of qualifying loss according to policy terms.
Companies should understand this before a dispute arises.
Check the AI and Technology Language
Insurance policies are evolving alongside AI.
Businesses should examine whether their EPLI policy contains provisions affecting:
- Algorithmic decision-making
- Privacy
- Biometric data
- Regulatory investigations
- Third-party vendors
- Cyber incidents
- Wage-and-hour matters
- Class actions
Don’t assume a policy purchased several years ago automatically addresses every modern AI hiring exposure.
Cyber Insurance May Also Become Relevant
An AI hiring platform handles valuable information.
Potential data can include:
- Names
- Email addresses
- Résumés
- Employment histories
- Assessment results
- Interview recordings
- Candidate profiles
A data breach involving the hiring platform may therefore trigger:
Cyber insurance issues
rather than—or alongside—EPLI.
One technology can create:
employment risk + privacy risk + cyber risk.
Insurance programs should be reviewed together.
Biometric Data Can Create Another Layer
Some recruiting technologies may analyze or collect information from:
- Video
- Voice
- Facial characteristics
- Behavioral assessments
That can potentially create additional privacy or biometric-law considerations depending on the jurisdiction and technology.
Before using such functionality, employers should understand:
what is actually collected
rather than assuming:
“It’s just an interview.”
Your Vendor Contract Matters
Suppose the AI vendor’s technology causes a serious problem.
Review whether the contract addresses:
- Indemnification
- Liability limits
- Insurance
- Data security
- Regulatory cooperation
- Audit rights
- Incident notification
- Data ownership
- Record retention
- Model changes
If your company carries:
$5 million EPLI
but your vendor contract limits the vendor’s liability to:
$25,000,
you may have a significant mismatch.
Maintain an AI Hiring Inventory
Businesses increasingly use AI without central management realizing it.
HR uses:
Tool A
Recruiting uses:
Tool B
Department manager uses:
Tool C
Staff download:
Tool D
Suddenly the company has four automated hiring technologies.
Create a simple inventory:
| Tool | Purpose | Owner | Candidates Affected | Last Review |
|---|---|---|---|---|
| Tool A | Résumé screening | HR | External applicants | Q2 2026 |
| Tool B | Candidate scoring | Recruiting | Graduate hires | Q1 2026 |
| Tool C | Interview support | Sales | Sales applicants | Q2 2026 |
You cannot govern technology you don’t know exists.
Audit the Entire Hiring Funnel
Don’t review only the final algorithm.
Examine:
Job advertisement
↓
Application
↓
Résumé screening
↓
Assessment
↓
Interview
↓
Candidate ranking
↓
Offer
↓
Rejection
At each stage ask:
Is automation involved?
What decision does it influence?
What data is collected?
Can a candidate request accommodation?
Who reviews the result?
Keep Appropriate Records
When a candidate challenges a hiring decision, you may need to explain what happened.
Useful records may include:
- Tool version
- Applicable hiring criteria
- Audit documentation
- Candidate notices
- Accommodation procedures
- Human review
- Vendor documentation
- Model changes
- Override records
Recordkeeping requirements can vary, so businesses should coordinate retention practices with employment counsel.
Don’t Let AI Write Unreviewed Rejection Messages
Automation can also create reputational problems.
Imagine a candidate receives:
“After reviewing your extensive experience, we have determined that your lack of relevant experience does not meet our requirements.”
They have:
15 years of relevant experience.
Clearly, nobody reviewed the message.
Even if this doesn’t create legal liability by itself, it can undermine trust and encourage candidates to question whether the entire process was legitimate.
Create a Human Escalation Path
Candidates should have a reasonable route for issues involving:
- Accommodation
- Technical problems
- Incorrect information
- Assessment accessibility
- Hiring-process questions
A completely automated hiring system with no practical human contact can make small problems harder to correct.
2026 AI Hiring Risk Checklist
Before deploying or renewing an AI hiring system:
- Identify every AI hiring tool in use.
- Determine which employment decisions each tool influences.
- Review federal discrimination requirements.
- Review applicable state and local AI laws.
- Determine whether bias audits are required.
- Review audit results.
- Establish candidate notices where required.
- Create an accommodation process.
- Provide alternative assessments where appropriate.
- Test candidate accessibility.
- Train recruiters.
- Maintain meaningful human oversight.
- Monitor hiring outcomes.
- Review rejection patterns.
- Document overrides.
- Review vendor contracts.
- Confirm vendor insurance.
- Review indemnification.
- Review candidate data collection.
- Review retention practices.
- Evaluate biometric-data exposure.
- Review EPLI coverage.
- Confirm applicant claims are addressed.
- Review regulatory-investigation coverage.
- Review cyber insurance.
- Reassess the system after major model updates.
Questions to Ask Your EPLI Broker
- Does our EPLI policy cover claims from job applicants?
- Are discrimination allegations involving AI covered?
- Is algorithmic hiring specifically excluded?
- How are class or collective claims handled?
- Are EEOC proceedings covered?
- Are state and local agency proceedings covered?
- Are regulatory investigations covered?
- Are defense costs inside or outside the limit?
- What retention applies?
- Are fines or penalties covered where legally insurable?
- Does third-party vendor involvement affect coverage?
- Are biometric-related claims excluded?
- Could cyber insurance respond to candidate-data incidents?
- Do we have overlapping EPLI and cyber coverage?
- Are our limits appropriate given our annual applicant volume?
Frequently Asked Questions
What is EPLI?
Employment Practices Liability Insurance is coverage designed for certain employment-related allegations such as discrimination, harassment, retaliation and wrongful termination, subject to policy terms.
Can an applicant sue over an AI hiring decision?
Potentially. Federal anti-discrimination laws can apply to hiring decisions involving AI just as they can to conventional hiring processes.
Is using AI to screen résumés illegal?
No. AI résumé screening isn’t inherently unlawful. But the employer must still comply with applicable employment-discrimination and other laws.
Is the employer responsible if a third-party AI tool discriminates?
Using a vendor does not automatically eliminate an employer’s legal exposure. Employers should evaluate both the technology and their own use of it.
Does New York City require AI hiring audits?
For covered automated employment decision tools, NYC Local Law 144 requires a qualifying bias audit and other requirements before use.
Does New York City require candidates to be notified?
Yes, covered use of an AEDT is subject to candidate or employee notice requirements.
Can AI discriminate against people with disabilities?
Potentially. The EEOC has warned that automated tools may screen out individuals with disabilities even when those individuals could perform the job with or without reasonable accommodation.
Does having a human recruiter eliminate AI liability?
Not automatically. If the AI determines which applicants reach human review, automated screening can still materially influence employment decisions.
Does EPLI automatically cover AI discrimination?
Don’t assume it does. Coverage depends on definitions, exclusions, claimant status, limits, retention and other policy provisions.
Should small businesses worry about AI hiring risk?
Yes, particularly if automated tools materially screen or rank applicants. A business doesn’t need thousands of employees before discrimination and compliance requirements become relevant.
Final Thoughts
AI can make hiring extraordinarily efficient.
It can:
read résumés in seconds
rank candidates instantly
schedule interviews automatically
answer applicant questions 24/7
and help recruiters manage applicant volumes that would once have required enormous teams.
But there is a fundamental principle employers shouldn’t forget:
Automating the decision doesn’t automate away responsibility.
The EEOC makes clear that existing federal employment-discrimination protections continue to apply when AI is used in employment.
And New York City’s AEDT rules demonstrate that some jurisdictions are imposing requirements specifically around automated hiring technology.
For employers, the better strategy is:
Know your tools → understand the data → test outcomes → provide accommodations → maintain human oversight → monitor vendors → document decisions → review EPLI.
The greatest AI hiring risk may not be that the machine makes an obviously terrible decision.
It may be that it quietly makes the same problematic decision thousands of times before anyone notices.