AI in HR: How Artificial Intelligence Is Changing Human Resources in 2027

6/9/2026

Artificial intelligence entered HR through the back door. Most organizations did not decide to adopt AI — their applicant tracking system added a ranking feature, their assessment vendor upgraded its scoring model, and suddenly automated tools were making decisions nobody had reviewed.

Quick answer: AI is now embedded across sourcing, screening, assessment, scheduling, employee support, and analytics. The legal exposure is not new law — existing anti-discrimination law applies fully to automated decisions — but a growing body of state and local regulation adds bias audit, notice, and transparency requirements. The employer remains liable for its vendor's model.

Where AI Is Actually Being Used

Function	Common Applications	Risk Level
Sourcing	Candidate matching, outreach personalization, job ad targeting	Moderate — targeting can produce disparate exposure to opportunities
Screening	Resume parsing, ranking, knockout questions, chatbot pre-screens	High — direct selection impact
Assessment	Game-based assessments, video interview scoring, skills testing	High — and video analysis carries additional disability and biometric concerns
Interview support	Transcription, summarization, structured question generation	Low to moderate — higher if scoring or recommending
Employee support	Policy chatbots, benefits Q&A, HR ticket routing	Low — but accuracy on leave and accommodation questions matters
Performance and retention	Flight risk prediction, performance analytics, promotion recommendations	High — and often invisible to the affected employee
Compensation	Market pricing, pay equity analysis, offer recommendations	Moderate to high — can encode historical inequity
Workforce planning	Demand forecasting, scheduling optimization	Moderate — scheduling algorithms can produce disparate impact

The Core Legal Principle

Existing employment law applies to AI-driven decisions without modification. Three consequences follow, and they are the whole framework:

1. Disparate impact liability applies. A neutral tool that produces a substantially different selection rate across protected groups requires the employer to show it is job related and consistent with business necessity — and even then, the plaintiff may show a less discriminatory alternative exists.
2. The employer is liable, not the vendor. "The vendor built it" is not a defense to a discrimination claim. Contractual indemnification may allocate cost, but it does not allocate liability.
3. Reasonable accommodation obligations apply to the tool. An assessment that disadvantages a candidate with a disability requires an accessible alternative, and candidates must be told how to request one.

The Disability Accommodation Problem

This is the exposure most employers have not thought through. AI tools can screen out qualified individuals with disabilities in ways that are invisible in aggregate statistics:

• Video interview analysis evaluating facial expression, eye contact, or speech patterns disadvantages candidates with autism, speech disabilities, facial differences, or neurological conditions.
• Timed assessments disadvantage candidates whose disability affects processing speed and who would be entitled to extended time as an accommodation.
• Game-based assessments may require motor skills or visual processing unrelated to the job.
• Chatbot screening may not be accessible to screen reader users.
• Employment gap penalties in resume screening disadvantage candidates who took medical leave.

Practical requirement: every automated assessment must be accompanied by a clear, prominent notice of how to request an accommodation, and an alternative evaluation method must actually exist — not be improvised when someone asks.

The Regulatory Landscape

Regulation is developing at state and local level, and the common elements are consistent enough to plan around:

Requirement Type	Typical Content
Bias audit	Independent audit of selection rates by sex, race, and ethnicity, conducted periodically, with results published
Candidate notice	Advance notice that an automated tool will be used, what it assesses, and how to request an alternative
Transparency	Disclosure of the data categories used and their source
Human review	A meaningful human decision-maker rather than fully automated rejection
Recordkeeping	Retention of tool outputs and decisions for a defined period
Biometric consent	Separate written consent where facial or voice analysis is used, under state biometric privacy laws

Biometric privacy laws deserve particular attention: several impose statutory damages per violation with a private right of action, which makes video interview analysis a disproportionately high-risk application.

A Governance Checklist

1. Inventory every tool. Include features embedded in systems you already own — most organizations discover they are using more AI than they knew.
2. Classify by decision impact. Tools that screen out or rank candidates get the most scrutiny; tools that summarize or schedule get less.
3. Demand vendor documentation. Training data description, validation methodology, adverse impact testing results, accessibility conformance, and audit history. Vendors who cannot supply this should not be used for selection decisions.
4. Run your own adverse impact analysis on your own applicant flow. A vendor's aggregate results do not tell you what the tool does with your candidate pool.
5. Require meaningful human review before any adverse decision. Rubber-stamping a machine recommendation is not human review.
6. Publish accommodation notices prominently and build an actual alternative process.
7. Address biometric consent separately where facial or voice analysis is involved.
8. Contract carefully. Audit rights, indemnification, data use restrictions, and notice of model changes. A model that changes mid-contract is a new tool.
9. Retain records of inputs, outputs, and decisions.
10. Review annually — regulation and models both change.

Where AI Genuinely Helps

The compliance framing can obscure real value. Applications with low risk and meaningful benefit:

• Job description drafting — with human review for essential-function accuracy
• Structured interview question generation, which improves consistency and reduces the ad hoc questioning that produces unlawful inquiries
• Policy and handbook drafting support, with legal review
• Employee self-service for routine questions, freeing HR for judgment work
• Scheduling and coordination
• Pay equity analysis, where statistical modeling genuinely outperforms manual review
• Sentiment and engagement analysis at aggregate level, with care around individual identification

The pattern: AI applied to preparation and analysis carries modest risk. AI applied to selection and evaluation of individuals carries substantial risk.

What AI Does Not Change

The judgment-heavy core of HR remains stubbornly human:

• Conducting an interactive process conversation
• Assessing witness credibility in an investigation
• Deciding whether a termination is defensible
• Coaching a manager through a difficult conversation
• Interpreting an ambiguous statute against unusual facts

These are exactly the areas where HR compliance training retains and increases its value. Automating the routine raises the proportion of HR work that is judgment — which raises the cost of poor judgment.

Frequently Asked Questions

Is it legal to use AI in hiring?

Yes, subject to existing anti-discrimination law and to jurisdiction-specific bias audit, notice, and transparency requirements.

Are we liable if the vendor's tool discriminates?

Yes. The employer is responsible for its selection procedures regardless of who built them. Contractual indemnity addresses cost, not liability.

Do we have to tell candidates we use AI?

Several jurisdictions require it, and it is advisable everywhere. Notice also creates the opportunity to offer accommodation, which reduces disability exposure.

What is a bias audit?

An assessment of the tool's selection rates across demographic groups, typically conducted by an independent auditor, with results retained and in some jurisdictions published.

Should we stop using AI in hiring?

Not necessarily. Well-validated tools can reduce inconsistency in human screening. The requirement is governance — inventory, validation, human review, accommodation, and documentation.

Judgment Is the Skill That Appreciates

As routine HR work automates, the remaining work is the work that carries legal exposure. Compliance depth becomes more valuable, not less.

The HR Generalist Certificate Program builds the foundation. For the highest-judgment areas, see the Certificate Program in FMLA, ADA, and PWFA Compliance and the Internal Investigations Certificate Program.

👉 Browse HR compliance training →

Additional resources: HR Compliance FAQ | Key HR Trends for Future HR Managers | People Analytics in HR