AI in hiring: govern the whole employment decision

A recruitment tool can affect someone before a hiring manager sees their application. Assess screening, ranking, interviews and promotion separately, then connect the applicable rules to notices, evaluations, accommodations and human review.

Last reviewed

Follow the candidate through each decision

Record the job location, affected population, employer, recruiter and vendor. Identify what the tool does: generate an advertisement, filter applications, score an assessment, infer characteristics, recommend a shortlist or determine an outcome. The tool's marketing label does not settle its legal scope.

Ask whether a human can meaningfully reconsider the result and whether a candidate can request an accommodation. Existing discrimination protections remain relevant alongside newer AI-specific duties. New York City's regulator expressly notes that fulfilling its audit requirements does not replace compliance with anti-discrimination law. DCWP guidance on overlapping obligations.

Final employment regulations · Effective October 1, 2025

California: employment discrimination rules already cover automated systems

The Civil Rights Council's automated-decision regulations clarify how the Fair Employment and Housing Act applies when covered employers and other entities use these systems. The approval notice describes prohibitions on discriminatory systems or selection criteria. These employment rules have applied since October 1, 2025. CRD final rulemaking record; approval and effective date.

This is a separate regime from California's privacy rules for significant automated decisions, whose Article 11 compliance date is January 1, 2027. Do not use the privacy timetable to postpone an employment-discrimination review. CCPA scope and staged deadlines.

Recommended practice: document job relevance, evaluate possible discriminatory effects, make accommodation routes usable, and retain the evidence needed to explain selection decisions. Agree with the supplier how changes and limitations will be disclosed. Purchasing a vendor tool does not resolve the employer's own obligations.

Local Law 144 · Enforcement began July 5, 2023

New York City: check the tool, audit and notice scope

Covered employers and employment agencies cannot use an automated employment decision tool without a bias audit conducted within the preceding year, public audit information and required notices. The rules concern tools that substantially assist or replace discretionary decisions in covered hiring or promotion. DCWP law and rules.

The city's guidance addresses location, including jobs associated with a New York City office, and screening before the final hiring decision. Required notices to covered New York City residents must precede use by ten business days. Do not treat a notice shown only after a rejection as the pre-use notice. DCWP applicability and notice FAQ.

An employer remains responsible for ensuring an audit exists even if the vendor arranges it. Check the auditor's independence, the assessed tool version, the data used and whether the audit can support your use. An audit does not certify that the tool is lawful or free of discriminatory effects. DCWP audit and responsibility FAQ.

Statutory AI provisions · Effective January 1, 2026

Illinois: distinguish the statute from withdrawn draft rules

Illinois Human Rights Act section 2-102(L) addresses discriminatory effects from AI in specified employment activities and the use of ZIP codes as proxies for protected classes. It also requires employee notice of covered AI use and directs the Department to make implementation rules. Current statute, section 2-102(L).

The state's June 26, 2026 rulemaking bulletin records withdrawal of the proposed AI employment amendments published in May. Those draft timing, format and recordkeeping details should not be presented as adopted requirements. Withdrawal of that proposal did not repeal the statute. This review does not establish a later final notice rule. JCAR withdrawal record.

Recommended practice: inventory affected employment uses and notices now; verify the current rulemaking position before fixing a notice template or retention schedule. Do not copy New York City's ten-business-day period into an Illinois policy as though it were an Illinois requirement.

Plan the other applicable layers

  • Colorado: employment is a consequential-decision category under the replacement ADMT law. Its main requirements begin January 1, 2027; the implementing rules remained drafts at this review. Colorado law and rulemaking status.
  • European Union: specified recruitment and worker-management uses appear in Annex III of the AI Act. The amended high-risk application date is December 2, 2027. Existing prohibitions and other applicable law can already matter: workplace emotion recognition is prohibited subject to the Act's medical or safety exception. Article 5; amended timetable.
  • United Kingdom: assess significant automated decisions under the amended UK GDPR, including meaningful human involvement and safeguards. The regulator's recruitment report identifies transparency, discrimination and misuse as review concerns; it describes voluntary engagement, not enforcement findings against all employers. UK requirements; Recruitment rewired.

These are selected regimes, not a complete employment-law inventory. Candidate location, disability protections, labor obligations and sector rules can require further assessment.

Make the procurement review useful after launch

Ask for a version-specific intended-use statement, input and output descriptions, evaluation scope, known limitations, accommodation support and a change-notification process. Where an audit is required, obtain its supporting report and confirm the scope with the responsible reviewers.

  1. Map the decision. Show every filtering and scoring step and the person responsible for the result.
  2. Connect the evidence. Keep evaluations, audit references, notices and their publication dates with the reviewed version.
  3. Exercise the exception route. Try an accommodation request, inaccurate input and a contested result using synthetic cases.
  4. Review changes. A new scoring threshold, model, job family or population should trigger a fresh scope and evidence check.

Liquid Learn supports service records, control owners and evidence references. It does not perform an independent bias audit, decide employment-law applicability or operate a candidate appeal process. Next editorial check: November 30, 2026, with particular attention to state rulemaking.

Connect the review to a real service

See how Liquid Learn connects AI service maps, accountable owners, controls and evidence. Or start with the 24-question health check for an immediate result and downloadable PDF roadmap.

Educational information, not legal advice. Applicability depends on jurisdiction, organizational role and use case. Confirm your obligations with qualified counsel. Liquid Learn supports governance work; using it does not certify or guarantee compliance.