Automated Employment Decision Tool

What counts as an automated decision system

The definitions differ by law, but they share a core idea: a computational process whose output (a score, ranking, classification or recommendation) is used to make, or substantially assist, a decision about a person's employment. In recruiting that includes:

• Resume screening and candidate matching that ranks or filters applicants
• Knockout logic that rejects applicants automatically
• Scored assessments, games and situational judgment tests
• Video or voice interview analysis and AI interview scoring
• Chatbots that screen candidates out based on answers
• Tools that set pay, assign shifts or flag employees for performance action

A calculator or spreadsheet that simply stores data is not usually in scope. The question is whether the output drives or heavily shapes a decision about a person.

The main US rules

• New York City Local Law 144 (enforced from July 5, 2023): uses the term AEDT. Employers and employment agencies may not use one for hiring or promotion decisions about NYC candidates unless it has had an independent bias audit within the past year, a summary of results is published, and candidates get notice at least 10 business days before use. See NYC Local Law 144.
• California Civil Rights Council regulations (effective October 1, 2025): clarify that using an automated-decision system that discriminates on a protected basis violates the Fair Employment and Housing Act, treat anti-bias testing as relevant evidence and require keeping ADS data for four years. Separate privacy regulations under the CCPA add notice and opt-out rules for automated decisionmaking technology from 2027.
• Colorado AI Act (SB 26-189, effective January 1, 2027): replaced the 2024 law (SB 24-205) before it took effect and regulates automated decision-making technology that materially influences consequential decisions, including employment, with notice, explanation, correction, human review and record-keeping duties for deployers and documentation duties for developers. See Colorado AI Act.
• Illinois HB 3773 (effective January 1, 2026): amends the Illinois Human Rights Act to bar AI use that has a discriminatory effect, prohibit zip codes as a proxy for protected classes and require notice to employees and applicants.

Federal anti-discrimination law also applies regardless of these: an employer is responsible under Title VII and the ADA for the outcomes of a tool it uses, even one built by a vendor.

Worked example

A logistics company uses a screening tool that scores applicants for driver roles from 1 to 100 and automatically advances those above 70. For a New York City warehouse, that is an AEDT under Local Law 144: the company commissions an independent bias audit comparing selection rates by sex and race/ethnicity, posts the summary on its careers site and adds a notice to the NYC job postings. For California applicants, it keeps four years of scores and inputs and tests for adverse impact. In Colorado, from 2027, it gives applicants a notice, sends rejected candidates an explanation of the tool's role within 30 days and offers human reconsideration. One tool, four sets of paperwork.

Human review is not an automatic exemption

Many employers assume that if a recruiter clicks "approve" on each AI recommendation, the law doesn't apply. That is risky. NYC's rule covers tools that "substantially assist" a decision, and Colorado covers tools that "materially influence" one. Real human-in-the-loop review means the reviewer sees the evidence, has time and authority to disagree, and sometimes does. A recruiter rubber-stamping 400 rankings an hour is not meaningful oversight.

How to govern these tools

1. Inventory every tool that scores, ranks or filters people, including features turned on inside your ATS.
2. Map which jurisdictions each one touches, based on where candidates and jobs are.
3. Ask vendors for documentation: intended use, training data, validation evidence and adverse impact testing.
4. Run or commission bias audits on your own outcomes, not just the vendor's sample.
5. Give candidates clear notice and an accessible alternative or accommodation route.
6. Keep inputs, outputs and decisions for the longest applicable retention period.
7. Review annually and after any model change.

Rules change quickly in this area and vary by state and country. This page is general information, not legal advice.