All regulations
New York City, USAEnforced since July 2023

NYC Local Law 144

The bias audit your automated employment decision tool needs

New York City Local Law 144 of 2021

Local Law 144 makes the audit something you do before you use the tool, not after. Screening candidates for a New York City job with automated software means three things. An independent audit within the last twelve months. The results posted on your site. And ten business days' notice to candidates before the tool runs.

Talk to us about an auditPlatform capabilities

What counts as an automated employment decision tool?

The city defines an automated employment decision tool as a computer-based tool that uses machine learning, statistical modelling, data analytics or artificial intelligence to substantially assist or replace a person's discretion in an employment decision.

The test is whether the tool produces a prediction or a classification, such as a fit score or a ranked shortlist, and weighs the inputs that drive that output. A resume screener, an assessment that scores candidates, a chatbot that filters applicants and a ranking model all qualify. A spreadsheet a recruiter reads and interprets does not.

Scope reaches further than the final hire. If the tool substantially helps assess or screen candidates at any stage of hiring or promotion, the audit and notice duties apply before you use it.

At a glance

Covered tools

Automated employment decision tools used to screen

Audit frequency

Annual, within the prior 12 months

Auditor

Independent third party

Core metric

Impact ratio against the four-fifths rule

Categories

Sex, race/ethnicity and the intersections

Enforcement

NYC Department of Consumer and Worker Protection

Who has to comply

  • Employers using an automated employment decision tool to screen candidates for a job located in NYC
  • Employment agencies operating in NYC
  • Employers hiring for fully remote roles where the only office is in NYC
  • HR-tech vendors whose customers carry the obligation

What the audit covers

  • Selection rate for each sex category
  • Selection rate for each race and ethnicity category
  • Impact ratio for every category against the most-selected group
  • Intersectional sex-by-race cross-tabulation
  • Optional exclusion of categories under 2% of the audit data, with the justification published
  • The number of applicants and scoring rate where the tool produces scores

Why single-axis testing is not enough

A tool can look balanced on sex and balanced on race while disadvantaging one specific sex-by-race group. Local Law 144 requires the cross-tabulation for exactly that reason, and it is where most disparities that matter surface.

We compute every combination in your data rather than only the largest cohorts, and disclose any group excluded for falling under the sample threshold.

Ten business days, before the tool runs

Candidates and employees living in New York City have to be told that an automated tool is in use, which job qualifications it assesses, and how to request an accommodation. That notice goes out at least ten business days before the tool touches their application.

You can serve it in the job posting, by mail or email, or by posting it in the employment section of your site and waiting the ten days out.

What non-compliance costs

Up to $500 for a first violation and up to $1,500 for each subsequent violation. Each day a non-compliant tool runs counts as a separate violation, and each failure to give notice counts separately again.

NYC Local Law 144 questions

A computer-based tool that uses machine learning, statistical modelling, data analytics or AI to substantially assist or replace human discretion in an employment decision. The test is whether it produces a prediction or classification, such as a fit score or a ranked shortlist, which weighs the inputs that drive that output.
Every year. An audit lasts twelve months from the day it was done. A tool you keep using needs a fresh one each year.
An independent auditor with no employment or financial relationship to the employer or to the vendor of the tool being audited. An audit run by the vendor's own data science team does not satisfy the law.
It applies when the tool is used for a job located in the city. DCWP treats a fully remote role as in scope where the employer's only office is in NYC. Where an employer has offices inside and outside the city, it becomes a fact-specific question turning on which location the role reports to. A role based at least part-time in an NYC office falls in scope, and an employment agency headquartered elsewhere still complies if the role meets one of those tests.
A summary of your most recent audit results. It needs the selection rates, the impact ratios and the date you began using the tool, which the rules call the distribution date. Put it on the careers part of your website. It stays up for at least six months after you last use the tool.
Take the selection rate for each category, meaning the share of applicants in that group who advanced. Divide each group's rate by the rate of the most-selected group to get its impact ratio. A ratio below 0.80 is the four-fifths threshold that signals adverse impact. The same calculation runs on sex, on race and ethnicity and on every sex-by-race combination. An auditor may leave out a category representing under 2% of the data used for the audit, provided the exclusion is justified in the published results.
It is the line on an application form telling you that software will help assess you. Local Law 144 gives you ten business days' notice before the tool runs, plus what the tool measures. Saying yes does not give anything up. You keep the right to bring a discrimination claim, and the employer still owes the yearly audit either way.
Local Law 144 requires employers to say how to request an alternative selection process or accommodation, so the request route has to exist and be published. The law does not force the employer to grant it. Separately, a disability-related accommodation request carries its own weight under the ADA and state law. Employers should expect these requests and decide in advance who handles them.
The law names employment agencies directly, so an agency screening candidates with an AEDT carries the duty itself rather than passing it to the client. In practice that means auditing the tools you run, giving notice to NYC candidates before you screen them and publishing your own summary. An agency headquartered outside the city is still covered when the role sits in NYC. Where a client's tool does the screening, work out in the contract who audits it and who publishes, because a gap there leaves both of you exposed.

Other regulations we audit against

Colorado SB 26-189

Colorado, USA

Notice, explanation and human review when automated technology materially influences a consequential decision.

Illinois HB 3773

Illinois, USA

Makes discriminatory AI in employment decisions a civil rights violation, and bans zip code as a proxy for protected class.

California FEHA AI rules

California, USA

Brings automated-decision systems inside the Fair Employment and Housing Act, with four-year records and testing that cuts both ways.

Get audited against NYC Local Law 144

One engagement can cover this and every other jurisdiction you operate in.

Talk to usHow we test
Automated employment decision tools: NYC Local Law 144