What Does an LL144 Compliance Checklist Actually Require in 2026?

What an LL144 Compliance Checklist Actually Requires in 2026

An LL144 compliance checklist is a structured review of whether an employer using an automated employment decision tool in New York City has satisfied the requirements of Local Law 144, formally known as Local Law No. 144 of 2021 and reflected in the New York City Administrative Code. The law took effect on January 1, 2023, and applies to covered employers hiring 20 or more employees for employment in an office in New York City when they use an automated employment decision tool to substantially assist or replace human discretion in a consequential employment decision. The checklist should cover the employer’s scope, the tool’s purpose, the annual independent bias audit, publication of the audit summary, candidate notice, language-assistance procedures, record retention, complaint handling, and the organization’s underlying nondiscrimination controls. A checklist does not replace legal analysis, and a completed form alone does not prove compliance. It is most useful when the employer documents what was reviewed, who performed the review, what evidence was collected, and how identified problems were corrected.

Also worth reading: What Should an Australian Small Business Include in an SMB HR Compliance Checklist in 2026? · What Is the AI Hiring Compliance Checklist Template for 2026 and How Do Employers Use It? · How Do Organizations Build a Reliable AI Recruitment Compliance Software Checklist?

Local Law 144 is narrower than many employers assume. It does not regulate every use of artificial intelligence in recruiting, and it does not automatically apply to every employee working in New York City. A tool used to schedule a shift, draft an internal announcement, or answer a general benefits question may be different from a system used to screen applicants, rank candidates, interview applicants, make hiring decisions, promote employees, or terminate employment. The organization therefore must first determine whether the employer has at least 20 employees hiring in New York City and whether the technology is an automated employment decision tool making or materially supporting a consequential decision. Employer size, hiring activity, employee location, tool functionality, and the degree of human involvement should be documented separately rather than treated as interchangeable facts.

The compliance review must also distinguish between a procedural requirement and a substantive one. The annual bias audit and public summary are affirmative obligations created by Local Law 144, but the law does not guarantee that an employer is free from discrimination. Existing federal, state, and city discrimination laws continue to apply. A tool can pass a narrowly designed audit and still create liability if selection procedures disproportionately exclude a protected group, if the employer ignores disparate-impact evidence, or if the employer uses an audit as an automatic defense. The right question is not whether a form is checked off, but whether the employer can show a defensible process for identifying, measuring, explaining, and correcting the risks associated with its system.

Covered Employers, Covered Decisions, and Scope

Before reviewing a checklist item, confirm the legal unit responsible for compliance. New York City guidance and the law address employers that hire 20 or more employees for employment in a New York City office. The 20-employee threshold is not a safe harbor for smaller companies that contract with a larger employer, use a vendor-controlled platform, or make decisions for a group organized under a different corporate structure. At the same time, a company that has 20 employees nationally is not necessarily covered if it does not hire employees for employment in a New York City office. Employment agencies and staffing firms can also face questions about how they use automated tools in the placement process, so the analysis should include outsourced recruiting and third-party platform arrangements.

The next question is what the system actually does. A consequential employment decision generally concerns candidates or employees being hired or promoted, or being terminated or laid off. Local Law 144 also addresses the use of automated tools in making decisions that affect the employment opportunity of a New York City applicant or employee. A system that generates interview questions, scores video interviews, evaluates a candidate’s personality, ranks applicants, or estimates future job performance should be examined even if a human manager clicks the final button. Human review must be meaningful rather than a rubber stamp. If the employer simply follows the system’s output without considering the underlying data, the tool may substantially replace discretion even when the employer has retained a nominal approval step.

This comparison is about evidence quality, not a ranking of legal obligations. An employer cannot transfer its statutory responsibility entirely to a vendor. A vendor may provide audit materials, but the employer should confirm whether the required audit is independent, whether it covers the employer’s actual use of the tool, and whether the published summary contains the required information. Likewise, a software dashboard can organize data, but it cannot by itself determine whether a tool qualifies under the law or whether the employer’s selection practices comply with discrimination rules.

The Annual Bias Audit and Public Summary

A central LL144 compliance item is the bias audit. A covered employer must conduct an audit of its automated employment decision tools to test for discrimination and bias against protected groups. The audit generally occurs at least annually, with the first audit completed within one year after the law became effective. The audit is not satisfied merely by asking the vendor whether the product is unbiased. The employer should understand the data used, the population tested, the selection rates or other impact measures, the statistical methods, the tester’s independence, and the reasons behind any identified disparities.

The test commonly discussed in connection with the four-fifths rule compares the rate at which a protected group is selected with the rate at which that group would be expected to be selected based on the relevant population. A selection rate below 80 percent of the comparison rate may raise a concern, but that ratio is not a universal safe harbor or a substitute for legal analysis. Statistical significance, small sample sizes, differences in job-related qualifications, the employer’s business needs, and the possibility of multiple selection stages all matter. An audit should not be reduced to a green, yellow, or red number without explaining what the number measures and what limitations apply.

The employer must publish a summary of the audit results, which ordinarily includes the type of tool audited, the audit date, the auditor’s identity, the audit’s substantive results, and information about data and methodology. The law and applicable guidance should be consulted for the exact publication timing and format. As a practical matter, the employer should build a publication calendar so that the summary is accurate, accessible, and not confused with confidential candidate records. A summary is not permission to publish names, interview answers, medical information, or other personal data. If the audit identifies a problem, the public summary should not be manipulated to hide it; the employer should describe the findings at the appropriate level and connect them to corrective action.

Candidate Notice, Language Assistance, and Human Review

Candidates must receive notice when an automated employment decision tool is used. The notice should be provided at the relevant stage, be understandable to the applicant, and identify when and how the tool is being used in the hiring process. A notice buried in a general privacy policy, delivered only after the candidate has been rejected, or written in language the candidate cannot reasonably understand is a weak compliance position. The notice should be integrated into the application workflow and tested with recruiters, hiring managers, vendors, and customer-facing recruiting teams.

Local Law 144 also requires a process allowing a candidate to request access to information about how the tool works and the types of data used to make the decision. That process must include a way to obtain an explanation and to request corrections when the applicant believes inaccurate information has been used. Language assistance must be available so that candidates can make requests and understand the response. The employer should not require a candidate to use a separate legal channel when the ordinary recruiting support process can resolve the request promptly. Records should show the request date, the employee or team handling it, the information or explanation provided, whether data was corrected, and the final outcome.

Human review should be real rather than ceremonial. Reviewers need sufficient time, relevant information, authority to depart from the tool’s recommendation, and training on how to document an exception. An employer should test whether managers routinely override the system, ignore it, or accept it without independent consideration. Those behaviors can reveal both compliance risk and operational weakness. If the employer’s written procedure says that a recruiter independently reviews every recommendation but actual practice allows recruiters to accept rankings in seconds, the written process is not enough. AI systems should also be monitored after deployment because performance changes when hiring volumes, labor markets, job descriptions, or the underlying data change.

Practical Steps for Building a Defensible Process

The first practical step is to create an inventory of every automated system used in recruiting and employment. Record the product owner, vendor, purpose, user population, decision stage, data sources, New York City exposure, and whether a protected group could be affected by the output. The inventory should include less visible tools, such as résumé-ranking systems, video-interview scoring, candidate-chat assistants that score answers, and internal promotion or termination models. A system may not be marketed as an AI product but still make or materially support a consequential decision. The inventory also helps the employer identify stale tools that are no longer used but remain connected to recruiting systems.

Next, collect the documents that demonstrate control. These materials may include a tool-use policy, approval records, vendor contracts, data-flow diagrams, audit plans, audit reports, publication links, candidate notices, request logs, training records, incident reports, and corrective-action plans. The employer should assign responsibility for each obligation. HR usually owns employment compliance, IT or security may own system data, procurement may own vendor oversight, legal may interpret the law, and the vendor may provide technical support. Shared ownership without a named accountable person tends to produce gaps. A short governance document is more useful than a long checklist that nobody updates.

The organization should then establish recurring reviews. At minimum, review new tools before deployment, conduct annual bias audits, update the public summary on the required schedule, and reassess the employer-size and New York City thresholds. A change to a scoring model, a new hiring office, a merger, a new vendor, or a shift from candidate screening to promotion management can alter the analysis. The employer should also test the candidate-request process at least periodically by submitting a sample request and checking the response time, accessibility, translation quality, and record completeness. The goal is to make compliance part of ordinary operations rather than a project that begins only after a regulator or claimant asks questions.

Common Mistakes That Create False Confidence

One common mistake is assuming that buying a bias-tested platform ends the employer’s obligations. A vendor’s certification or product documentation may support compliance, but the employer still needs to determine whether the tool is used as intended, whether the relevant hiring population is covered, and whether the required employer-specific audit and disclosures are in place. Another mistake is treating the four-fifths rule as a complete discrimination analysis. The ratio can identify a possible disparity, but it cannot explain whether the difference is justified, whether the sample is reliable, or whether the system is measuring the right employment outcomes.

Employers also make the mistake of treating human review as a safeguard without defining it. If a recruiter receives an automated score, has no time to investigate, and is evaluated on accepting the system’s recommendations, nominal review may not protect the employer. A second mistake is publishing a vague audit summary that omits the tool type, date, auditor, or substantive results. A third is sending notice only to rejected applicants. The process should be designed for the applicable stages of use, with reliable delivery evidence. A fourth mistake is failing to preserve records after a vendor changes or replaces the model, making it impossible to reproduce an earlier decision or explain what data was used.

Enforcement gaps do not mean that the obligations are optional. The National Law Review’s discussion of New York City enforcement limitations, Illinois disclosure requirements, and Connecticut developments illustrates why employers should distinguish between a legal remedy that has been tested and an obligation that may still be enforced differently or through evolving interpretations. A regulatory strategy should not depend on predicting which agency will act first. The safer approach is to document compliance while the law is clear and to obtain jurisdiction-specific advice when a tool, workforce, or business structure creates uncertainty.

Local Law 144, Illinois, Connecticut, and Federal Rules

New York City requirements are not the only employment-AI rules that may affect a national employer. Illinois employment discrimination rules now include notice and inquiry requirements connected to artificial intelligence used in recruiting, hiring, promotion, discharge, or other employment actions, with obligations generally becoming operative on January 1, 2026 under the relevant statutory and regulatory framework. Connecticut’s employment AI framework addresses employer notice, adverse-decision explanations, special-category data, and the treatment of applicants and employees, with a 20-employee threshold for much of the state law and enhanced notice duties for larger employers in relevant circumstances. Connecticut also retains separate job-applicant self-identification protections and restrictions concerning discrimination based on race, color, religious belief, sex, gender identity or expression, national origin, age, disability, or other protected status.

The federal Equal Employment Opportunity Commission’s 2024 guidance concerning discrimination in employment decisions and the use of artificial intelligence is also relevant to the same control environment. Its requirements, exceptions, and implementation dates should be checked for their current legal status in 2026, particularly because litigation or regulatory changes can alter the obligations. AI compliance should therefore be organized around durable principles rather than a single city rule: disclose meaningful use, assess disparate impact, protect applicant and employee data, permit meaningful review, preserve records, and investigate outcomes. A company that can explain these decisions in plain language is better prepared for a regulator, a candidate complaint, or a court challenge than one that relies on an unexplained vendor score.

Penalties, Costs, and the Timing of Action

The potential financial exposure is not limited to a voluntary audit. New York City’s law provides for civil enforcement by the Department of Consumer and Worker Protection, including penalties that can reach $2,500 to $5,000 for a first violation and $5,000 to $10,000 for a repeat violation, subject to the law’s exact terms. The law also provides a private right of action with potential damages commonly described as $500 to $2,500 per violation, and a violation may be treated as a separate violation for each day, subject to applicable legal limitations. A dispute over how those provisions apply should be analyzed by counsel rather than assumed from a summary of the law. Federal or state remedies may also be available when the same conduct creates discrimination, privacy, or consumer-protection risk.

Budgeting depends heavily on the existing program. A small employer with one low-risk screening tool might spend approximately $5,000 to $30,000 on an initial independent review, legal analysis, notice updates, and process documentation. A multi-tool recruiting platform with video, résumé, interview, and promotion components can cost substantially more, with annual audits, vendor fees, legal review, translations, and data remediation potentially ranging from $10,000 to well over $100,000. These are planning ranges, not government-set prices. Internal staff time is also a real expense, and an employer that delays may accumulate unresolved candidate complaints even if it ultimately avoids a penalty.

Action should begin before a new tool is launched, a New York City office opens, or a vendor reports a meaningful change. Organizations should prioritize tools that rank or reject applicants, use facial or voice data, evaluate video behavior, infer personality, or make termination recommendations. A company should not wait for a public enforcement announcement to determine whether its own records can support a defense. A 60- to 90-day initial project can identify systems, collect evidence, improve notices, and assign owners; ongoing monitoring is more useful than an occasional high-cost review. The objective is not perfect certainty, because no checklist can eliminate legal risk, but a traceable, current, and proportionate process gives the employer a stronger answer when asked how it handled algorithmic employment decisions.

The Best Compliance Approach for 2026

The best LL144 compliance checklist is one that combines legal scope analysis, independent testing, transparent disclosure, operational review, and human accountability. It should be tailored to the employer’s actual technology and workforce rather than copied from a generic list. The checklist should identify the applicable New York City threshold, define consequential decisions, name the responsible owner, confirm the annual audit, verify publication, test candidate notices and request handling, and document the employer’s response to any disparity. It should also connect to Illinois, Connecticut, and federal obligations when the employer operates across jurisdictions.

The strongest programs treat compliance as a continuing evidence system. They preserve audit reports, summaries, notices, access requests, training, model changes, and decision outcomes; they test whether reviewers can explain an automated recommendation; and they investigate when selection patterns differ across protected groups. They recognize that automation can improve consistency and reduce repetitive human work while also making errors faster and more scalable. No software feature or attractive annual report can repair a process that lacks accountability. A documented, independently supported, and regularly updated process is the most practical position an employer can take in 2026.