Appearance
Scope and applicability
Every other duty in Local Law 144 depends on one determination: is this tool an automated employment decision tool (AEDT) that is used in the city to make an employment decision? The statute defines the terms in § 20-870, the DCWP rules sharpen them in 6 RCNY § 5-300, and the DCWP FAQ (June 29, 2023) supplies the geographic and process boundaries. MFF-27 carries the determination as MRF-476.
The statutory definition — § 20-870
The term “automated employment decision tool” means 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. The term “automated employment decision tool” does not include 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.
And the decision the tool must feed:
The term “employment decision” means to screen candidates for employment or employees for promotion within the city.
Three elements have to hold at once: a computational process derived from one of the four named techniques; a simplified output; and use that substantially assists or replaces discretionary decision-making for screening. The rules define each.
The rules' tests — 6 RCNY § 5-300
| Element | What the rules say |
|---|---|
| Simplified output | A prediction or classification: a score, tag or categorization, recommendation, or ranking. It does not refer to the output of analytical tools that translate or transcribe existing text, such as converting a resume from PDF or transcribing a video or audio interview. |
| "Machine learning, statistical modeling, data analytics, or artificial intelligence" | Two limbs must both hold: the technique generates a prediction or a classification; and a computer at least in part identifies the inputs, the relative importance placed on those inputs, and, if applicable, other model parameters to improve accuracy. |
| "To substantially assist or replace discretionary decision making" | Three situations: to rely solely on a simplified output (score, tag, classification, ranking, etc.) with no other factors considered; to use a simplified output as one of a set of criteria where it is weighted more than any other criterion in the set; or to use a simplified output to overrule conclusions derived from other factors including human decision-making. |
| Candidate for employment | A person who has applied for a specific position by submitting the necessary information or items in the format required. |
A tool whose only output translates or transcribes existing text, such as converting a resume from PDF, does not issue simplified output for that function. A ranking model falls outside "substantially assist or replace" only if its output is not relied on solely, is not weighted more than every other criterion in the set, and is not used to overrule conclusions from other factors; a cutoff, an output weighted above every other criterion, or an output used to overrule other conclusions meets the element. The statutory exclusion list (junk email filter, firewall, antivirus software, calculator, spreadsheet, database, data set) applies only where the tool does not automate, support, substantially assist, or replace discretionary decision-making and does not materially impact natural persons; a spreadsheet used to rank and cut off applicants is not excluded by its label.
The process and geographic boundaries — DCWP FAQ
- Screening at any point counts (FAQ I.5); sourcing does not (FAQ I.6). Using a tool to scan a resume bank, conduct outreach, or invite applications is out of scope. Assessing people who applied for a specific position, at any point of the hiring or promotion process, is in scope.
- "Used in the city" (FAQ I.4). The law applies when the job location is an office in New York City at least part time; or the job is fully remote but the location associated with it is an office in the city; or the employment agency using the AEDT is located in the city (or, if it is not, one of the first two holds). Remote roles attached to a New York City office are in.
- Promotion is screening too. An internal mobility tool that ranks employees for promotion is an AEDT if it also meets the technique definition and one of the three "substantially assist or replace" prongs, and it then has its own notice channels; see Publication and notices.
The applicability test and the notice audience are different questions. Whether the law applies follows the "used in the city" test; who must be notified follows the residence of the candidate or employee (§ 20-871(b)).
Who is responsible
The employer or employment agency that uses the AEDT. Under the DCWP FAQ (V.1–V.2) the vendor that developed the tool is not responsible under the law, even if it commissions the bias audit. A vendor may have an independent auditor audit its tool and coordinate data collection, and several employers may rely on the same audit under the pooled-data conditions of 6 RCNY § 5-302(a); the deployer nonetheless remains responsible for ensuring that a compliant audit exists before use and that its own posting and notice duties are met.
Penalties — § 20-872 and 6 RCNY § 6-81
| Rule | Amount |
|---|---|
| First violation, and each additional violation occurring on the same day as the first (§ 20-872(a)) | Not more than $500 |
| Each subsequent violation (§ 20-872(a)) | $500 to $1,500 |
| DCWP penalty schedule (6 RCNY § 6-81) | $375 for a first violation ($500 on default), $1,350 for a second ($1,500 on default), $1,500 for a third or later violation |
Each day on which an AEDT is used in violation of § 20-871(a) is a separate violation, and each failure to provide a notice in violation of § 20-871(b) is a separate violation (§ 20-872(c)). Each day during a period of use on which the AEDT is used without a valid audit or posting is a separate violation (days on which it is not used are not); a hiring round that omits required notices can produce a separate violation for each failure under § 20-871(b)(1), (2), or (3), and under 6 RCNY § 6-81 each provision charged in the notice of violation is a separate violation. For schedule purposes, a second or later violation is a violation of the same provision by the same respondent, established by plea, decision, or settlement, within two years of the prior one. § 20-874 is a construction clause: the subchapter does not limit any candidate's or employee's right to bring a civil action, or the authority of the New York City Commission on Human Rights to enforce Title 8, so a Local Law 144 penalty does not exhaust the exposure of a discriminatory tool; the clause does not itself state that Local Law 144 creates a private cause of action.
How this maps in Modulos — MRF-476
MRF-476 (AEDT applicability determination) asks the project to determine and document whether the tool is an AEDT used in New York City: whether it is a computational process derived from the four techniques, whether it issues a simplified output, and whether that output substantially assists or replaces discretionary decision-making to screen candidates for employment or employees for promotion. The requirement carries the three prongs, the two-limb technique test, the "used in the city" test, and the candidate definition in its own text.
The requirement is satisfied by the reused control MCF-16 (Risk Tiering), the framework-agnostic control for classifying a system against regulatory criteria and recording the decision with rationale and decision-makers. For Local Law 144 the classification is binary: the tool is or is not an AEDT used in the city. MCF-16 is shared with FINMA AI Governance, the Singapore MGF for Agentic AI, and Microsoft Supplier DPR, and stays framework-agnostic; the Local Law 144 tests live in the requirement, not in the control.
One nuance the requirement states outright: the law imposes no record-keeping duty for this determination. The law requires the determination to be right, not to be written down. Documenting it with its rationale, and revisiting it when the tool, its output, or its role in the hiring or promotion process changes, is the practical way to evidence the boundary decision, and it is what the control records.
Where to go next
- The bias audit and its data — the duties that attach once the tool is in scope.
- Operationalizing Local Law 144 in Modulos — the full
MFF-27rollout.
Disclaimer
This page is for general informational purposes and does not constitute legal advice. Always verify against the current published text of N.Y.C. Administrative Code §§ 20-870–874 and 6 RCNY §§ 5-300–5-304 and consult qualified advisers.