The US scenario: AI in hiring meets a patchwork of rules
You are a TA or HR leader at a US company. You already use—or are evaluating—AI resume screening and structured interviews to handle volume. Then legal asks: "Is that tool an AEDT? Do we owe a bias audit? What do we publish, and to whom?" The challenge is not whether AI helps throughput; it is that the rules are a patchwork: federal EEOC guidance on adverse impact, NYC Local Law 144 bias-audit requirements, Illinois video-interview consent rules, and emerging state AI laws.
This guide helps you scope the issues, build a vendor due-diligence checklist, and design an audit trail so AI-assisted screening stays defensible. It is operational guidance, not legal advice—confirm specifics with employment counsel.
Executive summary
Split hiring into "collect → screen → structured interview → decide → retain/delete." Anchor screening to a versioned rubric so AI ranking and interview scoring are comparable and explainable. Keep human review with reason codes, document an adverse-impact review cadence, and where an AEDT applies, plan for an independent bias audit and candidate notices. Treat vendor "AI" claims as inputs to validate, not compliance you can inherit.
Know the regimes (scope with counsel)
| Regime | What it generally addresses | What to validate |
|---|---|---|
| EEOC / Title VII | Adverse impact and disparate treatment in selection procedures, including algorithmic tools | Whether your selection process shows adverse impact; job-relatedness and validation of criteria |
| ADA | Reasonable accommodation; tools that may disadvantage candidates with disabilities | Accommodation paths for assessments and video interviews |
| NYC Local Law 144 | Bias audits and notices for Automated Employment Decision Tools (AEDTs) used for NYC candidates | Whether your tool is an AEDT; who performs the independent audit; what you publish and notify |
| Illinois (AIVIA) | Notice, consent, and limits for AI analysis of video interviews | Consent flows and data handling for video interview analysis |
| State AI / privacy laws | Emerging rules on automated decisions, transparency, and consumer/candidate rights | Applicability by state of candidate residence; notice and opt-out mechanics |
Categories, thresholds, and definitions change. Treat this as a map of questions for counsel—not a determination that any law applies to you.
Audit-ready screening flow (bind data and decision versions)
AI provides ranking and scoring suggestions at C and D; humans review and record reason codes at E and F. An adverse-impact review and retention rules sit on the same chain at H, so you can answer "which rubric version drove this decision" and "what did our last impact review show."
Vendor due-diligence checklist (use with procurement and counsel)
| Area | Question to ask the vendor | Red flag |
|---|---|---|
| AEDT classification | Does your product meet the AEDT definition for our use case, and can you support a bias audit? | "We're compliant everywhere" with no scope detail |
| Bias audit support | What data do you provide for an independent audit (selection rates, scoring distributions)? | No access to score data or methodology |
| Explainability | Can rankings be explained and calibrated per role family? | Opaque scores presented as conclusions |
| Human-in-the-loop | How does your workflow enforce human review before adverse action? | Auto-reject with no documented review |
| Data handling | Retention, deletion, training-data use, and subprocessors? | Candidate data reused for model training without clear terms |
For evaluating tools on equal footing, see AI resume screening tools: an evaluation rubric, and for documentation patterns, regulated hiring documentation.
90-day rollout: one role family, measurable and defensible
Document the hiring state machine and a versioned rubric for one high-volume role; define KPIs (time-to-shortlist, pass rates).
Pilot AI screening and async video; confirm notices, consent, and retention with counsel; capture reason codes.
Calibrate scoring; run an adverse-impact review; if an AEDT applies, schedule the independent bias audit and publish required notices.
Expand to a second role; review audit completeness, retention hygiene, and accommodation paths before scaling.
Where MIND Interview fits (complement, not replace, your ATS)
MIND Interview is not an ATS and does not replace your system of record. It complements the screening and structured-interview layer: resume analysis for triage and ranking, and structured async video interviews with rubric-based scoring—so managers review evidence before live time. Status and score summaries can write back to your ATS per your retention policy. Structured rubrics and reason codes can make your own workflow easier to document.
No certification or audit claim: MIND Interview does not hold any regulatory certification and does not perform or constitute an independent bias audit. Where a bias audit or conformity assessment is required, it is the employer's responsibility and is carried out by an independent auditor you engage. Mention of any regime or third-party tool does not imply endorsement, compliance, or a determination that a law applies to you. Validate in your own due diligence and with counsel.
Related links
Enterprise AI recruitment: United States · Candidate data & AI resume screening · Screening tools evaluation rubric · Rubric calibration workshop. AI interview · Resume analysis · Pricing
Frequently Asked Questions
Key questions often raised by business leaders and HR teams:
Is this article legal advice?
No. References to EEOC guidance, NYC Local Law 144, and state laws are general information to help you scope issues with employment counsel and your privacy team. Confirm specifics with qualified legal counsel and current regulations.
Does NYC Local Law 144 apply to my company?
It generally applies to employers and employment agencies using an Automated Employment Decision Tool (AEDT) for candidates and employees in New York City—but coverage, the bias-audit requirement, and notice rules have nuances. Confirm scope with counsel before assuming you are in or out.
If our ATS already markets AI, do we still need a bias audit?
Possibly. Vendor marketing is not a substitute for your own obligations. Whether a tool is an AEDT, who runs the independent bias audit, and what you publish are fact-specific. Validate with the vendor and counsel, not slideware.
Does using AI for resume screening replace human review?
It should not. Decision-grade workflows keep human review with documented reason codes. AI provides ranking and summary suggestions; people make the hiring decision and can explain it.
How long should we keep candidate data and audit artifacts?
Per your retention policy and applicable law. What matters is auditable retention rules, expiry deletion/anonymization, and that backups and test environments are in scope.