What is an automated employment decision tool (AEDT)? The legal definition, examples, and how other laws compare
On this page
An automated employment decision tool (AEDT) is the term New York City's Local Law 144 uses for "any computational process, derived from machine learning, statistical modeling, data analytics, or artificial intelligence, that issues simplified output, including a score, classification, or recommendation, that is used to substantially assist or replace discretionary decision making for making employment decisions that impact natural persons" (NYC Admin. Code § 20-870). In plain terms: software that scores, ranks, tags or recommends candidates, where that output does most of the deciding. Resume rankers, scored assessments and tools that recommend who gets an interview can be AEDTs; spreadsheets, databases and interview transcription are not. Other laws regulate the same kind of tool under different names, with different thresholds: California's "automated-decision system," Colorado's "covered ADMT," Connecticut's "automated employment-related decision technology," California's privacy-law "ADMT" and the EU's "high-risk AI system," as of November 2026.
Not legal advice. Definitions below are quoted from the official texts linked in each row, as checked in November 2026. Whether a particular tool is covered depends on what it outputs and how your process uses that output, which only you can establish. Confirm with counsel.
What is an automated employment decision tool?
The New York City definition has four parts, and a tool must meet all four:
- A technique. Machine learning, statistical modeling, data analytics or AI. The city's rules narrow this to techniques that generate a prediction or classification and "for which a computer at least in part identifies the inputs, the relative importance placed on those inputs, and, if applicable, other parameters" (6 RCNY § 5-300).
- A simplified output. A score, classification or recommendation. The rules give four forms: a score, a tag or categorization, a recommendation and a ranking.
- Substantial assistance. The output is used "to substantially assist or replace discretionary decision making," which the rules define as relying solely on it, weighting it more than any other criterion, or using it to overrule other conclusions, including human ones.
- An employment decision. Screening candidates for employment or employees for promotion within the city.
The statute excludes "a tool that does not automate, support, substantially assist or replace discretionary decision-making processes and that does not materially impact natural persons, including, but not limited to, a junk email filter, firewall, antivirus software, calculator, spreadsheet, database, data set, or other compilation of data" (§ 20-870). How each test applies to interview tools feature by feature is worked through in Local Law 144 and AI note takers.
How do other laws define the same kind of tool?
The labels differ, and so do the thresholds. The phrase that decides coverage is in the third column.
| Term and law | Definition, as written | What brings a tool in | In force |
|---|---|---|---|
| Automated employment decision tool: NYC Admin. Code § 20-870 | As quoted above | Simplified output used "to substantially assist or replace discretionary decision making" | Enforced since July 5, 2023 |
| Automated-decision system: 2 CCR § 11008.1(a) (California FEHA) | "A computational process that makes a decision or facilitates human decision making regarding an employment benefit," which may use AI, machine learning, algorithms, statistics or other data processing | Making or merely facilitating a decision: the lowest threshold of the group | October 1, 2025 |
| Automated decisionmaking technology: 11 CCR § 7001(e) (California CCPA) | "Any technology that processes personal information and uses computation to replace human decisionmaking or substantially replace human decisionmaking" | Output used to decide "without human involvement," meaning no reviewer who knows how to interpret it, analyzes it with other information, and has authority to change the decision | Duties for significant decisions from January 1, 2027 |
| Covered ADMT: C.R.S. § 6-1-1701 (Colorado SB 26-189) | ADMT is "a technology that processes personal data and uses computation to generate output, including predictions, recommendations, classifications, rankings, scores, or other information that is used to make, guide, or assist a decision, judgment, or determination concerning an individual"; it is covered when "used to materially influence a consequential decision" | Output that is "a non-de minimis factor" and affects the outcome, "including by constraining, ranking, scoring, recommending, classifying" | Decisions on or after January 1, 2027 |
| Automated employment-related decision technology: Conn. P.A. 26-15, sec. 7 | "Any technology that processes personal data and uses computation to generate any output, including, but not limited to, any prediction, recommendation, classification, ranking, score or other information, that is a substantial factor used to make or materially influence an employment-related decision" | A "substantial factor": one that "meaningfully alters the outcome" | Notice duties for deployments from October 1, 2027 |
| Artificial intelligence: 775 ILCS 5/2-101(N) (Illinois Human Rights Act) | "A machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions," including generative AI | Any use "with respect to recruitment, hiring, promotion" and other listed decisions; no "substantial" threshold | January 1, 2026 |
| High-risk AI system: EU AI Act, Annex III point 4(a) | "AI systems intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates" | The provider's intended purpose, subject to the Article 6(3) exceptions | 2 December 2027 for Annex III systems |
Colorado's original 2024 act used a different term, "high-risk artificial intelligence system," defined as "any artificial intelligence system that, when deployed, makes, or is a substantial factor in making, a consequential decision" (SB 24-205). SB 26-189 repealed and reenacted that part of the statute before it took effect, so the term no longer governs Colorado hiring; see Colorado's AI law for employers. In the UK the question is framed as a decision "based solely on automated processing," explained in ICO guidance on AI in recruitment.
What is and is not an AEDT? Examples from the regulators
The clearest examples come from the regulators' own texts. These are their examples, grouped; they are not rulings on any product.
Likely inside the definitions
- Scores: "rating a candidate's estimated technical skills" (New York City's rules).
- Tags: "categorizing a candidate's resume based on key words, assigning a skill or trait to a candidate" (New York City).
- Recommendations: "whether a candidate should be given an interview" (New York City).
- Rankings: "arranging a list of candidates based on how well their cover letters match the job description" (New York City).
- Assessments: computer-based tests, questions, puzzles or games that make predictive assessments or measure skills, reaction time, personality, aptitude or "cultural fit" (California, 2 CCR 11008.1(a)(1)).
- Interview analysis: "analyzing facial expression, word choice, and/or voice in online interviews" (California).
- Targeting: directing job advertisements or recruiting materials to targeted groups (California); placing targeted job advertisements (EU).
Outside the definitions
- Transcription and translation: output from tools that "translate or transcribe existing text, e.g., convert a resume from a PDF or transcribe a video or audio interview" is not a simplified output (New York City).
- Ordinary software: spam filters, firewalls, antivirus, calculators, spreadsheets, databases and similar tools are excluded in New York City, California, Colorado and Connecticut, in California and Connecticut only so long as they do not themselves make or materially influence the decision.
- Summaries for a human reviewer: Colorado excludes "a tool used by an individual solely to summarize, organize, translate, draft, route, or present information" and actions that present information for human review without producing "a score, ranking, recommendation, classification, prediction, or other inference."
- Routine administration: Colorado excludes routine scheduling, administrative routing, communication of decisions and workflow management from consequential decisions.
- Sourcing before anyone applies: DCWP's FAQ says the New York City requirements do not apply to using a tool to scan a resume bank, do outreach or invite applications, because a candidate is someone who has applied for a specific position.
Why the same tool can be in or out
Most definitions are about use, not product category. New York City's three meanings of "substantially assist or replace" are all about weight: a ranking a recruiter glances at alongside the resume is different from a ranking that decides who gets a call. California's privacy rule turns on whether a qualified person reviews the output with other information and has authority to change the decision. Colorado's turns on whether the output is "a non-de minimis factor." The practical consequence is that a vendor's label settles nothing; your process does. Two checks are worth doing for every feature: what exactly does it output, and what happens to a candidate because of that output if no one looks further?
The term also differs from its neighbours. An applicant tracking system stores and routes applications and is not an AEDT for that, though a ranking feature inside it can be. An AI interviewer that scores answers may be one; an AI note taker's transcript is not, while any score it adds is assessed separately. A knockout question a person writes, with a fixed yes or no rule, does not use a technique in which "a computer at least in part identifies the inputs," which is how New York City's rules describe the covered techniques; the UK regulator reached a similar view for mandatory screening questions.
A copy-ready AEDT classification worksheet
AEDT CLASSIFICATION WORKSHEET (one per tool feature; example to adapt)
Tool, feature, version: [vendor / product / feature / version]
What it outputs: [transcript / summary / score / tag / rank /
recommendation / pass-fail]
Technique (vendor's answer): [ML, statistical model, LLM, fixed rules]
Does a computer set the inputs
or their weights? [yes / no / unknown: ask vendor in writing]
How the output is used: [sole factor / weighted most / one of several /
can overrule people / reference only]
Who reviews it, and do they
see every candidate? [role; yes / no]
Decision it affects: [advance / reject / interview / promote / pay]
Places in play: [job locations; candidate locations]
Likely position by law: NYC AEDT [Y/N/?] CA ADS [Y/N/?]
CCPA ADMT [Y/N/?] CO covered ADMT (2027) [Y/N/?]
CT (2027) [Y/N/?] EU high-risk [Y/N/?]
Duties if yes: [bias audit / notices / records / opt-out /
human review / logs]
Decided by / date / next review: [name, date, date]
What follows if a tool is covered?
In New York City: a bias audit within the past year, a public summary, and 10 business days' notice to candidates who live in the city (bias audit guide). In California: no discriminatory use, with anti-bias testing as evidence and four years of records (FEHA regulations), plus pre-use notice, opt-out and access rights under the CCPA ADMT rules from 2027. In Colorado and Connecticut, notices from 2027. Agencies carry these duties too (AI hiring laws for staffing agencies), and the records to keep are listed in AI hiring recordkeeping requirements. The full map is in AI hiring laws by state, and the vocabulary in the AI recruiting glossary.
Questions people ask
What does AEDT stand for?
AEDT stands for automated employment decision tool, the term New York City's Local Law 144 of 2021 uses for software built on machine learning, statistical modeling, data analytics or artificial intelligence that issues a simplified output, such as a score, classification or recommendation, used to substantially assist or replace discretionary decisions about hiring or promotion.
Is an interview transcription tool an automated employment decision tool?
Not for its transcripts. New York City's rules say a simplified output does not include output from tools that translate or transcribe existing text, such as transcribing a video or audio interview, and Colorado's 2026 law excludes tools used solely to summarize, organize or present information for human review. A feature of the same product that scores, ranks or recommends candidates has to be assessed separately, by what it outputs and how much weight your process gives it.
Is ChatGPT or another general chatbot an AEDT if a recruiter uses it to rank resumes?
It can be, depending on use. New York City's definition turns on the output and how it is used: relying solely on a ranking, weighting it more than any other criterion, or letting it overrule human conclusions all count as substantially assisting or replacing discretionary decision making. Colorado excludes general conversational technology only if it is not configured or intended for consequential decisions and an acceptable use policy prohibits that use.
Is an applicant tracking system an automated employment decision tool?
Storing and organizing applications is not: databases and similar software are excluded by name in the New York City, California, Colorado and Connecticut definitions. An ATS feature that scores, ranks or filters candidates and that recruiters rely on to decide who advances is a different matter and has to be assessed on its own.
What happens if a tool is an AEDT in New York City?
An employer or employment agency may use it to screen candidates for jobs in the city only if it had an independent bias audit within the past year, a summary of the results and the tool's distribution date is posted publicly, and candidates who live in the city get notice at least 10 business days before it is used, including the qualifications and characteristics it assesses and how to request an alternative process or accommodation.