Direct Answer: What Is the LL144 Audit Frequency in 2026?
If by LL144 you mean New York City Local Law 144, the answer as of September 24, 2026 is at least once per year for each covered automated employment decision tool, or AEDT, used to assist or replace discretionary hiring decisions. The rule is an annual minimum, not a requirement to audit every vacancy, every applicant, every month, or every quarter. The law also requires covered employers to publish the audit results and provide candidates with notice about the tool’s use. Those are separate obligations, and completing an annual report does not automatically satisfy the notice requirement.
Also worth reading: How Should Employers Build an AI Hiring Audit Checklist for 2026 to Ensure Compliance with Evolving State and Federal Regulations? · How Can Employers Conduct a Comprehensive AI Wage and Hour Audit in 2026? · What are the NYC automated employment decision tool audit requirements employers need to follow in 2026?
There is no single statutory filing date such as January 1 for every employer. The safer operating interpretation is to use a rolling 12-month schedule: if the last audit was completed on September 15, 2025, the next audit should be completed by approximately September 15, 2026, and certainly before the interval becomes longer than 12 months. If an employer cannot identify a prior audit, or if the previous audit was more than 12 months ago, the audit should be treated as due now. A calendar-year schedule can also work when it does not create a gap of more than 12 months and when the employer can document when the last audit occurred.
The annual frequency applies to the tool or system being used, not merely to the company’s general AI policy. If a company uses a resume-screening model, an interview-ranking system, and a separate candidate-search tool, each distinct system should be mapped to the rule and reviewed for its own audit obligation. Employers sometimes combine several features into one platform, and a combined audit can be defensible when the platform makes a connected hiring decision and the underlying components are sufficiently documented. What is not defensible is grouping unrelated tools simply to reduce the number of reports.
What LL144 Covers and What It Does Not Cover
Local Law 144 applies to employers and employment agencies that use a computational process derived from machine learning, statistical learning, or artificial intelligence to substantially assist or replace discretionary decision making in employment hiring. In practical terms, that can include software that ranks applicants, screens resumes, scores interview recordings, identifies candidates for interviews, or recommends which applicants should move forward. A tool does not have to make the final hiring decision by itself, and the presence of a human reviewer does not automatically remove the tool from the rule.
The core focus is recruiting and hiring, including recruiting applicants, selecting candidates for interviews, and making hiring decisions. A system used only to organize information may be outside LL144, but the boundary depends on how the employer configured and used the system. A vendor may describe a feature as a search, extraction, or administrative tool, while the employer’s actual configuration uses its output to screen people out or rank them for a discretionary decision. The employer should document the actual use rather than rely only on the product label.
The requirement took effect in New York City in 2023, and employers should not assume that a tool used for promotion, termination, scheduling, performance management, or compensation is automatically covered or automatically exempt. Those uses can raise separate federal, New York State, New York City, contractual, or internal policy questions. An AI inventory that labels every HR tool as either covered or not covered without reviewing the use case can therefore create more risk than clarity.
Why the Annual Audit Exists
The annual requirement reflects the fact that hiring systems operate in changing conditions. Applicant pools can change substantially from one year to the next, especially for seasonal jobs, campus recruiting, lower-volume roles, and jobs with specialized skill requirements. The employer may also change the scoring weights, the job description, the data sources, the vendor’s model, or the way recruiters interpret the output. A report that was accurate for a previous version may not describe the system actually operating in 2026.
A useful audit should explain the tool’s purpose, the data used, the categories analyzed, the methodology, and the results. New York City’s requirements call for examination of sex, race and ethnicity, and intersectional categories, along with information that allows a reader to understand how the audit was performed. Depending on the tool and the applicable rule, the report may include selection rates, impact ratios, counts of applicants, the qualifications or job categories used, and an explanation of limitations. The exact presentation is less important than making the population, time period, treatment of missing data, and calculation method understandable.
An annual audit is a compliance floor, not a finding that the tool is fair in every sense. Passing an LL144 audit does not prove that the employer has a legitimate, nondiscriminatory reason for every hiring outcome, and it does not prevent a candidate or employee from bringing a claim under another discrimination law. It also does not replace testing for disability accommodation, age, religion, sex, national origin, retaliation, privacy, or other obligations. The report should be used to identify needed changes, document testing, and support defensible decision making, not as a certificate that ends further review.
How to Run a Defensible LL144 Audit
Start with a precise inventory of the hiring technology. The inventory should identify the vendor, the product name, the model or configuration version, the hiring stage, the people responsible, the data received, the output produced, and the people affected by the output. It should also record whether a recruiter can override the result, whether the result automatically removes an applicant, and whether the system is used for more than one job family. A clear system boundary is essential because changing the audience or the decision being supported can change the appropriate test design.
Next, engage an independent auditor with the statistical and operational knowledge needed to evaluate the system. The auditor should be able to inspect the model version, the data pipeline, the decision thresholds, the methodology, and the results rather than merely repeat a vendor marketing statement. A vendor may provide raw data and technical documentation, but a firm that designed the tool should not automatically be treated as independent simply because it issued a report. The independence analysis should be documented, especially where one company supplies the software, performs the testing, and advises on compliance.
The audit should use a frozen, identified version of the system and a defined applicant or candidate population for the review period. The report should identify the number of records, the job categories, the time window, the protected-class fields used, the selection or exclusion events, the calculation rules, and the treatment of small groups or missing values. If the employer cannot obtain reliable demographic data, that limitation should be stated and addressed rather than silently removing the required analysis. After the results are reviewed, the employer should document corrective actions, retest material changes, publish the required information, and retain the report, underlying analysis, approvals, and version records.
How to Set the 2026 Audit Calendar
For 2026, the most important operational decision is to assign one responsible owner to the audit calendar. That owner should locate the last published report, confirm which tool and version it covered, and calculate the next due date. A company that completed an audit on January 10, 2026 should generally plan the next one before January 10, 2027, while a company whose last audit was on October 1, 2025 should act before the 12-month interval expires in October 2026. If the audit is already late, waiting for the next calendar year is not a sound response.
Set an internal deadline earlier than the legal deadline, such as 30 days before the anniversary, and use a second reminder 90 days before the due date. Calendar labels alone are not enough because they can conceal a long gap between a late-December audit and an early-January audit. A change in the model, scoring logic, vendor, applicant population, or job requirements should also trigger an internal review. The statute’s minimum is annual, but a fresh assessment is prudent when the prior report no longer describes the system in operation.
| Feature | Rolling 12-month schedule | Calendar-year schedule | Continuous monitoring plus annual report |
|---|---|---|---|
| Minimum cadence | One audit per covered tool every 12 months | At least one audit in each year, with no gap longer than 12 months | Continuous checks do not replace the annual audit |
| 2026 setup | Anchor the deadline to the last completed audit | Anchor reminders to January 1 and the prior report date | Automate change alerts, then commission the annual independent audit |
| Main benefit | Straightforward anniversary and fewer deadline disputes | Easy to explain in internal procedures | Earlier detection of drift, data breaks, or configuration changes |
| Main risk | A late audit can go unnoticed if ownership is unclear | Late-year or early-year gaps can create avoidable risk | Monitoring output may be mistaken for the required audit report |
A general AI inventory answers where a model exists, who owns it, and what it is intended to do. It is useful governance work, but it usually does not measure selection patterns by sex, race and ethnicity, or intersectional categories. A voluntary fairness assessment can be more detailed than the minimum LL144 report, and it may examine accuracy, error rates, accessibility, or business impact. That extra work can be valuable, but a voluntary assessment is not automatically a substitute for a required annual audit unless it contains the required elements and is published appropriately.
A federal or state discrimination review asks whether a hiring practice has an unlawful effect or whether reasonable accommodation and other legal duties are satisfied. It may consider a broader set of evidence, including interview questions, promotion decisions, accommodation requests, and historical patterns. LL144 focuses on a narrower and more specific duty for covered hiring tools. Employers that treat the two reviews as interchangeable risk missing either the city’s publication and notice rules or the broader discrimination analysis.
Software can help collect model versions, schedule deadlines, store reports, and flag configuration changes. It can also help a compliance team compare a current report with the prior year’s report. However, automation does not itself make the auditor independent, establish a defensible methodology, or replace legal judgment about whether the tool is used in a covered way. The best technology reduces administrative work while leaving the substantive testing, interpretation, publication, and remediation with accountable people.
Common Mistakes That Create 2026 Risk
One frequent mistake is interpreting once per year as a calendar-year slogan rather than a deadline tied to the last completed audit. Another is assuming that a vendor’s annual report is sufficient without reviewing the covered system, version, population, and independence of the reviewer. A report can be technically impressive yet fail to describe the tool the employer actually used. Employers also make the error of treating the audit as a notice substitute, publishing a report but failing to give candidates the required notice about the AEDT.
Another error is leaving protected-class data out of the analysis because it is inconvenient or difficult to collect. The proper response is to document the data limitation, examine available data, and determine whether additional data collection or a revised methodology is needed. A company should not manufacture demographic data, infer sensitive attributes without a defensible basis, or simply declare a result fair because the system produced no protected-class field. The law is focused on actual hiring effects, not on whether an employer can avoid measuring them.
A final mistake is treating the completed report as the end of the process. If the audit finds a disparity or a weak data condition, the employer should document the response, test any correction, and determine whether the change is material enough to require a new report. Vendor model updates, workflow changes, and revised job qualifications should be logged. The annual report is a point-in-time record of a particular system, not a permanent defense against later changes or other claims.
Cost, Pricing, and the Role of Compliance Software
New York City does not impose a fixed government filing fee for an LL144 bias audit, and the employer is responsible for obtaining and paying for the work. A narrow project involving one stable tool, a clean data extract, and a modest applicant population may receive vendor quotes around $5,000 to $15,000. A project with several models, several job families, messy historical data, multiple jurisdictions, and a requirement to test remediation can exceed $25,000 and sometimes reach the six figures. These are market ranges, not statutory rates, and the final price depends heavily on data preparation and the depth of testing.
The largest hidden cost is often remediation. If a model must be rebuilt, thresholds must be changed, a vendor must be replaced, or recruiters must revise how the output is used, the cost can exceed the audit itself. Compliance software may be priced by user, module, or volume, with annual subscriptions ranging from hundreds to several thousand dollars for smaller deployments. That software can reduce document handling and missed deadlines, but it should be treated as a workflow aid rather than a guarantee of legal compliance.
For a 2026 program, request a written scope that states the tools included, the audit frequency, the independence arrangement, the data fields, the report contents, the publication method, and the price for annual refreshes. A low quote that excludes data cleaning or a narrowly defined model may create more expense later. A higher quote may also be poorly targeted, so employers should compare deliverables rather than select the cheapest report or the most expensive dashboard.
When to Act and What to Do Now
Act immediately if the last LL144 audit is older than 12 months, the last report cannot be located, the tool changed after the report was issued, or the employer cannot show that the report was published. An acquisition, a new recruiting platform, a vendor migration, or the introduction of automated interview scoring can change the covered system even when the company’s brand and job descriptions remain the same. The employer should also act if candidates were not told that an AEDT was used, because a missing notice obligation is separate from a missing annual audit.
For a 24 September 2026 review, record the last audit date for every covered tool, identify the current model version, assign an owner, and place the next deadline on a tracked calendar. Keep the final report, the methodology, the underlying data description, the auditor’s independence statement, remediation decisions, publication evidence, candidate notices, and model-change logs. These records make it easier to explain not only that an audit occurred, but that the employer understood the tool, tested it, responded to results, and maintained the required public information.
The shortest accurate answer is therefore simple: LL144 requires at least one bias audit per year in 2026, with publication and candidate notice, and a rolling 12-month schedule is the most defensible way to manage the deadline. The annual report is a baseline obligation, not permission to stop reviewing a changing hiring system. If the tool affects a decision outside ordinary recruiting, or if the employer is uncertain whether a vendor feature substantially assists hiring discretion, it should obtain New York City employment counsel’s view rather than relying on a product label or an old report.