New York Law Journal, “‘AI Made the Decision’ Is Not a Defense: What NYC Employers Need to Know About the City’s AI Enforcement Position,” by Keith J. Gutstein, Esq.,11-2-26

Kaufman Dolowich’s Keith J. Gutstein, Chair of KD’s Labor and Employment Law Practice Group and Co-Managing Partner of the firm’s Long Island office, authored an insightful article in the New York Law Journal discussing New York City’s enforcement position on AI-assisted employment decisions and what employers need to know about their legal obligations when using artificial intelligence in the workplace.

Read the full article below: 

‘AI Made the Decision’ Is Not a Defense: What NYC Employers Need to Know About the City’s AI Enforcement Position

NYC says employers remain liable for unlawful AI-assisted decisions. Audits and human review help, but don’t replace legal compliance.

By Keith Gutstein, Kaufman Dolowich LLP

Nov. 2, 2026

As artificial intelligence becomes increasingly embedded in employment decisions, employers face a fundamental question: Who bears legal responsibility when an algorithm contributes to an unlawful result? New York City has reaffirmed that employers remain responsible for complying with applicable anti-discrimination and worker-protection laws when they use automated systems. Reliance on an algorithm or a third-party technology provider does not, by itself, excuse a violation.

The city underscored that message on Oct. 1, when the Mayor’s Office, the New York City Department of Consumer and Worker Protection (DCWP), Commission on Human Rights (CCHR) and Taxi and Limousine Commission (TLC) issued a joint enforcement policy statement addressing the use of artificial intelligence, algorithmic tools and automated systems. The agencies emphasized that businesses cannot avoid responsibility under existing laws merely because technology generated or influenced a decision.

Although the statement does not itself create new legal requirements, it signals a coordinated enforcement position as employers increasingly rely on automated technologies in workplace decision-making.

For employment counsel, the announcement raises important questions about the interaction between existing laws, the limits of regulatory compliance and the evidence necessary to defend AI-assisted employment decisions.

The announcement also comes amid broader scrutiny of artificial intelligence in New York City, including an Oct. 5 City Council oversight hearing examining the risks posed by AI and potential regulatory safeguards.

Existing Laws, New Enforcement Emphasis Central to the existing legal framework is the New York City Human Rights Law (NYCHRL), which prohibits discrimination and discriminatory harassment based on protected characteristics, as well as retaliation for engaging in protected activity. Employers using automated systems remain subject to these prohibitions, including when seemingly neutral technologies produce discriminatory outcomes.

An AI-powered screening tool, for example, may rely on historical employment data that reflects past disparities. Even without discriminatory intent, the resulting recommendations could raise disparate-impact concerns if they disproportionately disadvantage members of a protected group. The NYCHRL expressly recognizes disparate-impact claims under Administrative Code §8-107(17).

More broadly, the NYCHRL extends beyond hiring and promotion, to compensation, discipline, termination and other terms and conditions of employment.

The city’s worker-protection laws create additional exposure. Automated systems used to calculate wages, establish schedules or administer protected time off must comply with applicable requirements. An algorithm that improperly denies protected time off or, for an employer covered by the city’s Fair Workweek Law, generates schedules inconsistent with applicable requirements, may result in violations notwithstanding the absence of direct human involvement.

Beyond these substantive requirements, employers should also take note of the city’s broader enforcement posture. The enforcement announcement follows Mayor Zohran Mamdani’s Sept. 3 Executive Order No. 20, which directs DCWP, CCHR and TLC, in coordination with the Deputy Mayor for Economic Justice, to develop directed investigation procedures for workplace violations and other predatory practices even when no complaint has been filed.

The Limits of Local Law 144 Compliance

Against this backdrop, New York City’s Automated Employment Decision Tool (AEDT) law, commonly known as Local Law 144, provides a more targeted regulatory framework.

Local Law 144, which took effect Jan. 1, 2023, with enforcement beginning July 5, 2023, generally requires employers and employment agencies using covered automated tools in hiring or promotion decisions to ensure that a bias audit by an independent auditor has been conducted within the preceding year, publish required audit information and provide applicable notices.

Local Law 144 covers specified computational processes derived from machine learning, statistical modeling, data analytics or artificial intelligence that issue a simplified output, such as a score, classification or recommendation, used to substantially assist or replace discretionary decision making in hiring or promotion.

The implementing rules cover reliance on the output alone, weighting it more heavily than any other individual criterion, or using it to overrule conclusions reached from other factors, including human judgment.

Consequently, requiring a manager to approve an algorithmic recommendation does not necessarily remove the technology from the law’s scope.

Local Law 144’s requirements are limited to qualifying tools used in hiring and promotion. Automated systems used solely for discipline, termination, scheduling or compensation generally fall outside that framework but remain subject to applicable anti-discrimination and worker-protection laws.

Compliance with Local Law 144 should not be mistaken for a determination that an automated tool is nondiscriminatory. DCWP’s guidance states that the law does not require specific actions based on bias-audit results and reminds employers that other anti-discrimination laws remain applicable.

Moreover, the required audit calculations address sex, race/ethnicity and intersectional categories but do not assess every characteristic protected by the NYCHRL. A completed audit, therefore, does not necessarily establish compliance with the NYCHRL.

Vendor Algorithms and Employer Liability

Separate concerns arise when employers rely on third-party technology. Many use commercially available platforms for recruiting, performance management and workforce administration without detailed knowledge of the underlying algorithms.

Yet reliance on vendor assurances also does not eliminate an employer’s legal obligations. DCWP’s Local Law 144 guidance assigns responsibility for ensuring completion of required bias audits to employers and employment agencies, even when vendors coordinate those audits.

The October enforcement statement reinforces the broader principle that regulated entities remain accountable for compliance with laws applicable to technologies they deploy.

Although the statement creates no new discovery or documentation requirements, its reaffirmation of existing obligations has practical implications for how employers investigate, explain and defend AI-assisted decisions.

When an employer relies on an automated recommendation for an adverse employment action, questions may arise about the information considered, criteria applied and extent to which decision-makers independently evaluated the recommendation.

Plaintiffs may seek information about system inputs, the weight assigned to particular factors, bias testing, technology changes and instances in which decision-makers departed from its recommendations. Employers unable to identify the relevant inputs, the system’s role and the basis for the ultimate decision may have difficulty explaining the challenged action and responding to discovery.

A vendor’s assertion that its technology is proprietary may further complicate discovery involving algorithmic inputs, decision-making criteria and records of system performance.

In a disparate-treatment case, an employer’s ability to identify the factual basis for an AI-assisted decision and explain the role of the automated recommendation may become important in evaluating its stated legitimate, nondiscriminatory business reasons and any allegation of pretext.

Disparate-impact claims require a different analysis. Under Administrative Code §8-107(17), counsel must consider the challenged practice, its asserted impact, the applicable statutory defenses and any available less discriminatory alternative.

Employment counsel should evaluate whether vendor agreements provide sufficient access to relevant documentation, audit findings and information concerning material changes to the system. Agreements should also address cooperation with investigations, litigation and regulatory inquiries.

Employers should understand what information a system evaluates, whether its outputs can be independently verified and how potential inaccuracies or discriminatory effects are identified and addressed.

Counsel should therefore be involved when employers select and implement these systems, not only after an AI-assisted decision is challenged. Early legal review can identify regulatory obligations, potential evidentiary gaps and contractual limitations that may be more difficult to address after a dispute arises.

Human Oversight as a Legal Safeguard

Meaningful human review requires more than a supervisor’s approval. A manager who routinely accepts automated recommendations without understanding their basis may complicate an employer’s ability to explain or defend a challenged decision.

For hiring and promotion, employers should document the job-related criteria considered, the role of automated recommendations and the factual basis for the ultimate determination.

Disciplinary and termination decisions warrant particular scrutiny. Before acting on an algorithmic assessment, decision-makers should verify the underlying facts, consider relevant workplace policies and evaluate whether protected leave, reasonable accommodations or other legally relevant circumstances affected the reported conduct.

For example, an automated attendance system may identify an employee as having excessive absences without adequately distinguishing unprotected absences from legally protected leave. Reliance on that output without further examination could create exposure.

Documentation should demonstrate the substance of the review, including any discrepancies identified, additional information considered and reasons for accepting or rejecting an automated recommendation.

Employers should also establish procedures for challenging inaccurate outputs and preserving relevant records, subject to applicable retention requirements and litigation holds.

These safeguards cannot guarantee immunity from liability. They can, however, help establish that employment decisions resulted from a reasoned assessment rather than unquestioning reliance on technology.

The Continuing Responsibility of Employers

New York City’s October announcement does not prohibit employers from using artificial intelligence or impose a new comprehensive AI compliance regime. Its significance lies in reaffirming that technological innovation does not displace established legal obligations.

Employers and their counsel should assess automated systems throughout the employment relationship, determine which regulatory requirements apply and ensure that appropriate oversight accompanies their use.

As AI becomes more deeply integrated into workplace decision-making, disputes may increasingly turn on how employers evaluated algorithmic recommendations, what information informed their decisions and whether those decisions complied with applicable law.

The technology may change how employment decisions are reached. It does not change who must answer for them.

Keith J. Gutstein is chair of Kaufman Dolowich’s Labor & Employment Law Practice Group and co-managing partner of the firm’s Long Island office.

“Reprinted with permission from the Nov. 2, 2026 issue of “New York Law Journal” © 2026 ALM Global Properties LLC, trading as Centellic. Further duplication without permission is prohibited contact 877-256-2472 or asset-and-logo-licensing@alm.com.  All rights reserved.