NYC Local Law 144 and automated hiring tools
The first law anywhere to require a published bias audit before an automated tool may help decide who gets hired. It is narrower than its reputation, and the narrowness is where employers get caught.
New York City Local Law 144 requires an employer or employment agency using an automated employment decision tool to screen candidates for a job or promotion in New York City to have that tool independently bias-audited within the previous year, to publish a summary of the audit results, and to notify affected candidates at least ten business days in advance. Enforcement is by the Department of Consumer and Worker Protection, with penalties per violation and each day of continued use treated separately.
Last reviewed
The short version
- The trigger is a tool that substantially assists or replaces discretionary decision-making — not merely any software in the hiring stack.
- The audit must be independent and no more than a year old, and its summary must be publicly posted on the employer’s site.
- Candidates get at least ten business days’ notice, including the job qualifications and characteristics the tool assesses.
- The duty sits with the EMPLOYER, not the vendor. A vendor can supply the audit and the data; it cannot discharge your obligation.
- CalHire never auto-rejects, which changes the analysis but does not by itself remove a tool from scope. Assume you are in scope and check.
What the law actually turns on
Three definitions decide whether any of this applies to you. They are worth reading slowly, because most of the confusion about this law lives in them.
- Automated employment decision tool (AEDT)
- A computational process derived from machine learning, statistical modelling, data analytics or artificial intelligence that issues a simplified output — a score, classification or recommendation — used to substantially assist or replace discretionary decision-making about employment. A keyword filter a recruiter wrote by hand is generally not one; a model that ranks a pipeline generally is.
- Substantially assist or replace
- The DCWP rules give three cases: relying solely on the output, using it as one factor weighted more than any other, or using it to overrule a human conclusion. This is the test doing the real work. A score a human genuinely weighs alongside other evidence is treated differently from one that decides the shortlist.
- Bias audit
- An impartial evaluation by an independent auditor that, at minimum, calculates selection rates and impact ratios by sex and by race/ethnicity category, and by the intersection of the two. It must use historical data from the tool’s actual use where available; test data may be used only where historical data is insufficient, and the summary must say so and why.
What compliance actually involves
In the order you have to do them, because the notice obligation catches employers who did the audit but scheduled it wrong.
- 1
Establish whether the tool is in scope
Ask what the output is and what it decides. If a score orders your shortlist and the order is what you act on, treat it as in scope regardless of what the vendor calls it.
- 2
Commission an independent bias audit
Independent means the auditor was not involved in using, developing or distributing the tool and has no financial interest in the outcome. You will need selection-rate and impact-ratio figures by sex, by race/ethnicity and by their intersection.
- 3
Publish the summary before use
On your own website, publicly, with the distribution date of the tool and the date of the audit. It stays up while you use the tool and must be no more than a year old.
- 4
Give candidates ten business days’ notice
Before use, stating that an automated tool will be used, the job qualifications and characteristics it will assess, and — on request within the window — the data source and retention policy. Ten BUSINESS days, which is two calendar weeks plus holidays.
- 5
Keep the record
The audit, the published summary, the notice you sent and the date you sent it. If the tool changes materially, you are looking at a new audit rather than an amended one.
Where CalHire sits, stated plainly
CalHire produces a composite score that ranks a pipeline. A ranking is a simplified output used in a hiring process, so the honest position is to assume it can fall within the definition rather than to argue it out of scope — and to design as though it does.
One structural fact changes the risk profile: no code path on CalHire can reject a candidate. There is no score threshold that ends an application, no auto-advance, and no configuration that turns one on. A human makes every progression decision and is recorded as having made it. That does not remove a tool from the law’s scope on its own, but it puts you squarely outside the first of the three "substantially assist" cases and gives you evidence against the other two.
What we provide is the material an audit needs and the notice an employer has to send: selection rates and impact ratios by stage, exportable with dates; the decision record naming the human who made each call; and candidate-facing notice text you can adapt. The free adverse impact calculator on this site runs the same four-fifths arithmetic an auditor will, if you want to see where you stand before commissioning one.
What we do not provide is the audit. An audit CalHire performed on CalHire would not be independent, and a vendor-run "audit" offered as satisfying this law is one of the clearer signs that a vendor has not read it.
Primary sources
The rules have been amended since the law passed and the enforcement guidance has moved. Read the regulator rather than any summary, including this one.
- Automated Employment Decision Tools — rules, FAQ and enforcementNYC Department of Consumer and Worker Protection
- Uniform Guidelines on Employee Selection Procedures — tests and selection proceduresU.S. Equal Employment Opportunity Commission
What this page does not do
It is not legal advice, and it is not a determination that any particular tool is or is not in scope. That question depends on how YOU use the output, which is a fact about your process rather than about the software.
It does not make CalHire compliant on your behalf. Local Law 144 places the duty on the employer or employment agency. We can hand you the data and the records; the audit, the publication and the notice are yours.
CalHire has not been independently bias-audited to date, and we do not claim to have been. Our current position, including what we can and cannot yet show, is on the bias audit disclosure page rather than implied here.
Passing a bias audit is not a finding of fairness. The four-fifths ratio is a screening heuristic; clearing it leaves intersectional effects and small-group distortions entirely intact.
This covers New York City only. Other jurisdictions have their own automated-decision rules, and the trend is toward more of them rather than convergence.
Questions people actually ask
Does Local Law 144 apply to us if a human makes the final decision?
Can our vendor’s bias audit satisfy the requirement?
How long is a bias audit good for?
What has to be in the candidate notice?
Is CalHire an AEDT?
What happens if we get this wrong?
Related
Where this connects to the rest of the platform.
Hiring in the United States
Federal anti-discrimination law applies everywhere, and a growing set of state and city rules add bias-audit, notice and disclosure duties on top.
AI governance, bias auditing and the audit trail
Adverse-impact analysis on the four-fifths rule, a hash-chained audit trail, candidate appeals, and DSAR handling. The evidence exists before anyone asks for it.
The EU AI Act and hiring systems
Employment AI is classified high-risk under Annex III. That brings human oversight, logging, transparency and data-governance duties — and the obligations fall on the employer deploying the system as well as the provider building it.
Read the reasoning
The evidence and the argument behind what is on this page.
NYC Local Law 144: a practical compliance checklist
Local Law 144 requires an annual independent bias audit, a published summary, and advance notice to candidates. What each obligation means operationally.
ReadWhat counts as an automated employment decision tool?
The AEDT definition turns on whether a tool substantially assists or replaces discretionary decisions. Where the line falls, and the cases teams get wrong.
ReadWhat an AI bias audit actually measures
A bias audit is an outcome analysis, not a code review. What impact ratios are, what an audit cannot tell you, and how to read a summary you have been handed.
ReadSee a verified pipeline for one of your roles
Post a role free and review anonymous, skill-ranked candidates. No card, no sales call to get started.