Recruiters · problem and fix
NYC local law 144 bias audit requirements for AI screening: a recruiter's Mac checklist
If you use AI to rank, score, or filter applicants for a job in New York City, NYC Local Law 144 probably applies to you. It requires an independent bias audit, a public summary of the results, and notice to NYC-resident candidates 10 business days before you use the tool. This page covers each rule, the math behind the four-fifths check, and what you can do on your Mac this week.
The short answer
- The law covers automated employment decision tools that substantially assist hiring decisions, including screening.
- You need an independent bias audit. You can rely on it for one year from the date it was conducted.
- Post a public summary of the latest results on your website, with the audit date, data source, and applicant counts.
- Give NYC-resident candidates notice 10 business days before the tool is used. Then check your own numbers with the four-fifths rule.
What bias audit and notice for ai screening costs you
Enforcement and penalty exposure
Violations can bring civil penalties. This page does not list amounts, because we did not confirm them against the city's own FAQ. Anti-discrimination laws still apply on top of this law, so a biased tool can create a claim even when the audit paperwork is in order.
A hiring decision you cannot explain
If a candidate asks why they were screened out, you need an answer. An old audit, or one from an auditor with ties to the vendor, will not give you that answer. Your records should show what the tool did and what a person decided.
A summary you cannot publish
Without a current audit, you cannot post the required summary. The gap sits on your public website, and candidates and reporters can see it. A missing summary is easy to notice and hard to explain.
Rework when the dates slip
Redoing a notice or an audit late costs time and attention from people who already have full calendars. This page has no verified figure for how much time that takes. Put the audit date and the notice date on a calendar before a requisition opens.
Trust with candidates
Applicants who never hear that a tool scored them may feel shut out. The notice gives them a path to ask for an accommodation. Skipping it turns a routine step into a complaint, and it can start with one applicant who compares notes online.
Use is common, which is not the same as fair
SHRM's 2025 Talent Trends survey found 43% of respondents use AI for HR work, mostly recruiting, and nearly 90% of those users say it saves time or improves efficiency. Those figures show the tools are popular. They do not show the tools are fair, so the audit still matters.
Why it happens
Language models learn patterns from names
A University of Washington study tested three LLMs as resume rankers. The models favored white-associated names 85% of the time, against 9% for Black-associated names. For male-associated names, the figures were 52% and 11% for female-associated names. The study used more than 550 real-world resumes and more than 500 job listings across nine occupations. It tested ranking, not your hiring funnel, so read it as a warning about the tool type.
Bias moves into education and other proxies
A 2025 NAACL industry-track study by Iso et al. found that newer LLMs have reduced bias tied to explicit gender and race attributes. Bias tied to educational background remained significant. A model can look clean on names and still rank people in a biased way through school or degree signals. Check the proxies the tool uses, not only the obvious fields.
Loose criteria leave the tool to fill gaps
A tool scores what its setup tells it to score. If the job criteria are vague, the tool fills the gaps with patterns from past data or generic signals. That makes a rank harder to explain to a candidate, and harder for an auditor to test. Tight, written criteria give you something to check.
Small pools hide real gaps
A small applicant pool can make an impact ratio swing from one round to the next. The EEOC warns that small numbers can mislead in both directions. Teams often read a clean ratio as proof, when the count is too small to tell anything.
Rubber-stamp review
When a recruiter accepts the top results without reading the rest, the tool is making the decision. The law covers tools that substantially assist or replace decisions. Rubber-stamp review does not take a tool out of scope, and it removes the human reasoning that would catch a problem.
Unclear ownership between teams
Recruiters often assume the vendor handled compliance. Vendors often assume the employer will publish the summary and send the notice. Legal teams may assume the recruiting team has the list of tools. The gap between those assumptions is where notices get missed.
How to fix it
These work with or without any app.
Map every tool that scores or ranks
List each tool in your hiring stack, including resume parsers, scorecards, assessment platforms, and applicant tracking filters. For each one, note whether it scores or ranks and whether a person relies on the output. Mark the tools that touch NYC candidates. Keep the list in a Google Sheet or a Numbers file that anyone on the team can edit.
Ask each vendor the audit questions in writing
Ask for the latest bias audit, its date, its data source, the applicant counts, and the auditor's name. Ask the auditor to confirm independence against the DCWP criteria. Send the questions from Gmail so you have a dated record. If a vendor cannot answer, pause NYC use until you get answers.
Build an impact ratio sheet yourself
Export applicant counts by category from your tracker. Put one group per row. Use columns for applicants, selected, selection rate, and impact ratio. If applicants are in column B and selected are in column C, the selection rate in D2 is =C2/B2. The impact ratio in E2 is =D2/MAX($D$2:$D$9), with the range matching your rows. Select the formulas and fill them down with ⌘D. Flag any ratio under 0.8. This is a first check, not a replacement for the audit.
Read small groups with their counts
When a group has few applicants, do not treat a flag as proof, and do not treat a pass as proof either. The EEOC says a difference over 20% may not show adverse impact when very few people are selected. Put the count next to every ratio in your sheet, so the person reading it sees the sample size.
Draft the notice before you need it
Write one template that says the tool is used, lists the qualifications or characteristics it assesses, and explains how to request a reasonable accommodation. Store it as a Gmail draft or a doc. Fill in the tool name and dates for each role. Have your HR or legal lead approve the wording once, so it does not change every round.
Put the deadlines on two calendars
Set one reminder for the notice date, counted back 10 business days from the first day the tool runs. Set a second reminder for one year after each audit date. Apple Reminders works for this on a Mac. Create both reminders when the role is set up, not the week it opens.
Require a written reason for each decision
Ask the recruiter to add one line for each candidate who moves forward or stops. Store those notes with the tool output. The notes show that a person made the call. They also help you spot patterns that the audit numbers may miss.
Test removing name fields before you rely on it
Ask your vendor whether names, photos, or addresses can be removed from scoring. The University of Washington study shows names can shift rankings in LLM tools. No verified study shows this step reduces bias in your pipeline, so test it on a sample and compare the results before you change the process.
Repeat the ratio check every round
Run the impact ratio sheet after every hiring round, not only when the audit expires. The sheet shows drift between audits. Keep the old versions with dates, so you can show what changed and when. Follow your employer's rules on candidate records when you save these files.
NYC Local Law 144 says an employer or employment agency that uses an automated employment decision tool to screen NYC candidates or employees must do three things. It needs an independent bias audit, it must publish a summary of the results, and it must give notice before the tool is used. The city's Department of Consumer and Worker Protection (DCWP) publishes the official FAQ.
What counts as an automated employment decision tool
The law uses the term automated employment decision tool, or AEDT. The city's FAQ defines it as a computer-based tool that uses machine learning, statistical modeling, data analytics, or AI. The tool must substantially assist or replace discretionary decision-making. The law says screening counts as an employment decision.
Ask three questions. Does the tool score, rank, or filter candidates? Does a person rely on its output to move people forward or stop them? Are any candidates or employees in NYC? If the first two answers are yes and the third is yes, treat the tool as covered until counsel says otherwise.
Do not assume a tool is too simple to count. A filter that scores or ranks applicants can fall under the definition, because the definition covers data analytics. Read the vendor's documentation. Look for words like score, rank, match, or recommend, and treat them as a signal to check.
The three duties and the one-year audit clock
| Duty | What it must include | Timing |
|---|---|---|
| Independent bias audit | Impartial evaluation by an independent auditor, with selection or scoring rates and impact ratios by sex, race or ethnicity, and intersections | Rely on it for one year from the audit date |
| Public summary | Latest audit results, audit date, data source, applicant counts, selection or scoring rates, impact ratios | Post on your website or behind an active hyperlink |
| Candidate notice | Tool use, job qualifications or characteristics assessed, and instructions for requesting a reasonable accommodation | Give 10 business days before the tool is used |
The law does not require a specific action based on the audit results. DCWP says you must still follow federal, state, and NYC anti-discrimination laws. A passing audit is not a shield. Treat it as one input to your legal review, not as the end of the question.
What the independent bias audit must cover
The audit must be an impartial evaluation by an independent auditor. It calculates selection or scoring rates and impact ratios across sex, race or ethnicity, and intersectional categories. An intersectional category combines two traits, such as sex and race or ethnicity. Ask whether the audit includes those combined groups.
Before you accept any audit, write down the auditor's name and the audit date. Check that name against your company, your vendor, and the tool's distribution history. Keep that note with the audit file. It takes a few minutes and saves a hard conversation later.
Impact ratios and the four-fifths rule, explained
An impact ratio compares one group's selection rate with the highest group's rate. The selection rate is the number selected divided by the number who applied. To get the ratio, divide each group's rate by the top group's rate. A ratio of 1.0 means the group matches the top group exactly.
The EEOC uses a rule of thumb called the four-fifths rule. If a group's selection rate is less than four-fifths, or 80%, of the highest group's rate, the EEOC treats it as a substantially different rate of selection. The EEOC says this is not a legal definition. Treat it as a warning light, not a verdict.
Here is a worked example with invented numbers, so you can check the math. Say the top group advances at a rate of 50%. Four-fifths of 50% is 40%. Any group advancing below 40% falls under the rule of thumb. A group that advances at 30% has a ratio of 0.6, which is under 0.8, so it gets a flag.
Small groups can mislead in both directions. The EEOC notes that a difference of more than 20% may not show adverse impact when the number selected is very small. Large numbers can show adverse impact from smaller gaps, if those gaps are statistically and practically significant. Read every ratio beside its counts.
43%
of surveyed HR professionals use AI for HR work, mostly for recruiting
SHRM: Uncertain Economy Adds to Recruitment Challenges (Talent Trends 2025)85% vs 9%
Times LLM resume rankers favored white-associated names over Black-associated names in one study
University of Washington News: AI bias in resume screening by race and gender80%
Four-fifths threshold: a group selection rate under 80% of the top group rate is treated as substantially different
EEOC: Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform GuidelinesHow to write and send the NYC candidate notice
The notice goes to NYC-resident candidates and employees. It must say that an AEDT is used. It must name the job qualifications or characteristics the tool assesses. It must explain how to ask for a reasonable accommodation. Plain language works best. A candidate should understand it on one read.
Count the 10 business days backward from the first day the tool runs. Skip weekends and public holidays. If a holiday falls inside that window, send the notice earlier. Put the send date on your calendar as well as in your mail schedule, so two reminders agree.
If you send from Gmail, use the schedule option, then check the date and time before you close the window. The step-by-step version is in How to schedule an email to send later in Gmail. Keep a copy of the sent notice with its date in your hiring file.
Common mistakes with NYC Local Law 144
- Treating the audit as a one-time task. It is good for one year from the date it was conducted.
- Publishing a summary without the applicant counts, data source, or audit date.
- Sending the notice on the day the tool starts. It must go out 10 business days earlier.
- Accepting a vendor audit without checking whether the auditor has ties to the vendor or your company.
- Missing NYC residents in a national applicant pool. The notice duty still applies to them.
- Using the four-fifths rule as a pass or fail test. It is a rule of thumb, and small samples mislead.
Watch for changes to the rules. Some secondary sites say a narrower replacement rule is coming. We did not confirm that against the city's own text. Check the DCWP FAQ before you plan a hiring cycle that runs past any change.
Where Lazur helps, and where it does not
Lazur is a voice assistant for your Mac. Hold ⌥ ⌘ and ask about anything on screen, and Lazur answers out loud. It can explain a term in the DCWP FAQ, point at a cell or field, guide you one click at a time, and draft text for you to check.
Lazur does not run an audit, approve an auditor, or decide who is covered. It does not click buttons or operate other apps for you. It will not send the notice or post the summary. Those steps stay with you, your counsel, and your auditor.
Be careful with candidate data. Lazur sends audio and the screenshot to AI providers to answer you, and they are not stored. Follow your employer's rules on candidate records before you show a resume or applicant list on screen. If your employer bans AI on those records, do not hold the shortcut on those screens.
Start with the tool list. Every other step on this page depends on knowing which tools score or rank NYC candidates. Lazur has a 3-day Pro trial with no credit card if you want to test it on this work. Pricing is on the pricing page.
Faster with Lazur
Where Lazur helps with bias audit and notice for ai screening
Hold ⌥ ⌘ and say what you want. Lazur sees your Gmail window. Lazur draws a circle and a label around the button or menu you need.
- 1
Hold ⌥⌘ and talk
- 2
Lazur draws a circle and a label around the button or menu you need.
- 3
You stay in Gmail. Nothing to copy into a chat window.
More things to say in Gmail
- “What does substantially assist mean in this DCWP FAQ?”
- “Where is the impact ratio column in this sheet?”
- “Walk me through the impact ratio formula, one step at a time.”
- “Draft a candidate notice for a role that starts on a Monday.”
- “Look up any change to the DCWP FAQ and summarize it.”
- “Remind me on the notice date I counted for the NYC role.”
Free for 3 days, then $18/month. Download for Mac
You say“Lazur, where is the impact ratio column?”
Lazur · points at itLazur circles the impact ratio column in your sheet and labels it, so you can check each value against 0.8.
What recruiters say to Lazur for this
Each one is a single hold of ⌥ ⌘.
“What does substantially assist mean in this DCWP FAQ?”
Lazur reads the FAQ on screen with you and explains the term in plain words, out loud.
Explains what's on screen
“Where is the impact ratio column in this sheet?”
Lazur draws a circle and a label around the column on your screen, so you find it fast.
Points at it
“Walk me through the impact ratio formula, one step at a time.”
Lazur guides you one click at a time and waits for each click before it moves on.
Walks you through it
“Draft a candidate notice for a role that starts on a Monday.”
Lazur drafts the text. You check every line against the DCWP FAQ before you send it.
Drafts it
“Look up any change to the DCWP FAQ and summarize it.”
Lazur runs research in its own Chrome tabs and returns a summary and a report. Check any change against the city's own page.
Researches in the background
“Remind me on the notice date I counted for the NYC role.”
Lazur adds an Apple Reminders item for the date you choose. You still count the business days yourself.
Handles the small stuff
A plan you can start this week
- Day 1
List every tool that scores or ranks
Open a Google Sheet and add a row for each tool in your hiring stack. Mark whether it scores, ranks, or filters, and whether a person relies on the output. Mark which tools touch NYC candidates.
- Day 2
Ask vendors for their audit papers
Email each NYC-relevant vendor for the latest bias audit, its date, its data source, the applicant counts, and the auditor's name. Save the replies under one Gmail label so they stay together.
- Day 3
Check auditor independence
Compare the auditor's name against the vendor, your company, and the tool's history. Look for past work on the tool or a financial tie. If you are unsure, ask the auditor to confirm in writing.
- Day 4
Build the impact ratio sheet
Pull applicant counts by category from your tracker. Enter selection rates and impact ratios with formulas. Flag any ratio under 0.8, and write the sample size beside each flag.
- Day 5
Draft the candidate notice
Write the notice with the tool name, the qualifications or characteristics it assesses, and accommodation instructions. Have counsel or your HR lead review the wording before anything goes out.
- Day 6
Set the notice date
Count back 10 business days from the first day the tool runs, skipping weekends and holidays. Schedule the notice for that date, or earlier if a holiday sits in the window.
- Day 7
Post the summary and set the audit reminder
Publish the audit summary with the audit date, data source, applicant counts, selection rates, and impact ratios. Set a reminder for one year from the audit date.
Questions people ask
Does NYC Local Law 144 apply if my company is not based in New York City?
It can. The notice duty covers NYC-resident candidates and employees, so check whether anyone in your pool lives in NYC. The scope questions for out-of-state employers are not settled on this page. Check the DCWP FAQ and ask counsel before you decide.
Is a bias audit the same as the four-fifths rule?
No. The bias audit is an independent evaluation with impact ratios by sex, race or ethnicity, and intersectional categories. The four-fifths rule is an EEOC rule of thumb for spotting a substantially different selection rate. The EEOC says it is not a legal definition.
How long is a bias audit good for?
DCWP says employers and employment agencies can rely on a bias audit for one year from the date it was conducted. After that, you need a new audit before you keep using the tool for NYC candidates.
Do I have to change the tool if the audit shows a gap?
The law does not require a specific action based on the audit results. Federal, state, and NYC anti-discrimination laws still apply, so a gap can still create legal risk. Talk to counsel before you decide what to change.
Can I run the four-fifths check myself in Excel or Numbers?
Yes, for a first check. Calculate each group's selection rate, then divide each rate by the top rate. Flag any result under 0.8 and note the counts beside it. Your independent audit still has to cover the legal requirements.
Do candidates have to get the audit results themselves?
The public summary goes on your website or behind an active hyperlink. The candidate notice covers tool use, the qualifications or characteristics assessed, and how to ask for an accommodation. Keep both on file.
What if a candidate asks for a reasonable accommodation?
The notice must include instructions for requesting a reasonable accommodation. Send the request to your HR or accommodation contact and follow your employer's process. Do not decide the request alone in the recruiting inbox.
Is Lazur an automated employment decision tool?
Lazur is a voice assistant for your Mac. Its listed features are explaining, pointing, guiding, dictating, drafting, research, and everyday tasks. They do not include scoring or ranking candidates. Your hiring software is a separate question, and this is not legal advice.
Sources
- Automated Employment Decision Tools: Frequently Asked Questions · NYC Department of Consumer and Worker Protection (DCWP) · checked 2026-11-09
- University of Washington News: AI bias in resume screening by race and gender · University of Washington · checked 2026-11-09
- Uncertain Economy Adds to Recruitment Challenges (SHRM Talent Trends 2025) · SHRM · checked 2026-11-09
- EEOC: Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform Guidelines on Employee Selection Procedures · U.S. Equal Employment Opportunity Commission · checked 2026-11-09
- Evaluating Bias in LLMs for Job-Resume Matching: Gender, Race, and Education · arXiv (NAACL 2025 Industry Track) · checked 2026-11-09
Fix bias audit and notice for ai screening with Lazur, free for 3 days
Hold a key and ask. Lazur sees your screen, points at the button and talks you through it, in any Mac app. No credit card.