NYC Local Law 144

An AEDT is a computational process derived from machine learning, statistical modeling, data analytics, or artificial intelligence that issues a simplified output (score, classification, ranking, or recommendation) used to substantially assist or replace discretionary decision-making in employment decisions. DCWP rules clarify that substantially assist includes relying on the output as the primary factor or using it to overrule human conclusions.

The definition captures more than purpose-built HR software. A general-purpose AI tool becomes an AEDT in function when a recruiter asks it to score or shortlist candidates and the output drives the decision. Organizations need an inventory of every place AI output touches hiring, including informal chatbot use, because the audit and notice obligations attach to the use, not to the vendor's product category.

An AEDT may not be used unless it has undergone a bias audit by an independent auditor no more than one year before use. The audit must calculate selection or scoring rates and impact ratios across sex and race/ethnicity categories, using the employer's historical data where available.

For organizations using AI on candidate data, this means every screening tool needs a current, independent audit on file, refreshed annually, and the data pipeline feeding the tool must be clean enough to compute impact ratios by demographic category. It also means an ad hoc AI workflow a recruiter invented cannot be audited, because nobody knows it exists: discovering actual AI use in hiring is a precondition of compliance.

Before using an AEDT, the employer or agency must publicly post on its website a summary of the most recent bias audit results, including selection rates and impact ratios, and the distribution date of the tool.

Publication turns bias metrics into public, citable evidence. Organizations should assume plaintiffs' counsel, journalists, and regulators read these postings and compare them across years. That raises the stakes on data quality: what candidate data enters the tool, how categories are recorded, and whether the posted numbers can be reproduced from the underlying records.

Candidates and employees who reside in NYC must be told, at least 10 business days before the AEDT is used, that an AEDT will assess them, which job qualifications and characteristics it will use, and how to request an alternative selection process or accommodation. Information about the tool's data sources and retention policy must be available on request.

Notice obligations assume the employer knows which tools will run before screening starts, which is incompatible with recruiters improvising with unapproved AI mid-process. The data-source and retention disclosure also forces the employer to know what its vendors keep. Candidate resumes and personal details pasted into a consumer chatbot sit outside every one of these commitments: no notice was given, no retention policy applies, and no audit covered it.