Seyfarth Synopsis: On November 4, 2022, New York City’s Department of Consumer & Worker Protection (“DCWP”) held a public hearing on its proposed rules implementing Local Law 144 of 2021 (“LL 144”). LL 144, which regulates employers’ use of automated employment decision tools, including artificial intelligence and other “computational processes,” is scheduled to go into effect January 1, 2023. While the record was to close by midnight, DCWP did not provide a firm date on when the proposed rules would be finalized.

On November 4, 2022, the DCWP held a hearing to provide the public with an opportunity to comment on the Department’s proposed rules implementing LL 144. For further reading about the passage and history of LL 144, our past coverage can be found here.

The highly anticipated hearing was attended by over 275 participants and saw over 60 minutes of testimony from various constituencies, such as academics, employers, law firms, consultants, employee advocates, and technology vendors. Seyfarth submitted written comments and partners Annette Tyman and Robert T. Szyba testified at the hearing. While a transcript of the hearing is expected in the coming days, the following key themes emerged.

  • Automated Employment Decision Tool (AEDT) definition: Several commenters, including Seyfarth’s Annette Tyman, recommended further clarification to be provided in defining “AEDT,” such as identifying with greater specificity the “computational processes” that will require a bias audit. Commenters pointed out that the definitions and explanations provided in the proposed rules discuss generalized concepts and use broad terminology, making it very difficult for the employer community to know with certainty which “computational process[es]” are covered, and which are not covered.

In contrast, several other commenters called for even broader definitions, testifying that the proposed rules defined covered AEDTs too narrowly. For example, one commenter explained, by narrowly interpreting the phrase “to substantially assist or replace discretionary decision making,” the rules permit a loophole that would allow employers to argue that certain AEDTs are not covered where the employer does not assign enough weight to the outcome or result of the AEDT.

  • Independence of Bias Audits: Multiple commenters brought up the limited guidance under the proposed rules addressing the level of independence needed when conducting a bias audit. Commenters made several suggestions where additional clarity would be helpful including: (1) greater guidance on who should be allowed to conduct the independent bias audit, and (2) how the audits should be conducted. Some commenters recommended that the bias audits should be handled internally by either the employer who actually uses, and is thus familiar with the application of the AEDT in their workplace. Alternatively, commenters suggested that the developers of such tools are best positioned to conduct bias audits. Other commenters, however, recommended that bias audits should be conducted by entities unaffiliated with the use or development of such technology. There were also differences expressed on how to conduct bias audits with some suggesting that a universal standard should be developed and utilized, whereas others argued that a one-size-fits-all approach would be problematic given each AEDT’s unique design and implementation. Other commenters recommended that the audit criteria be clearly defined and include the industry and legal standards.
  • Notice Requirements: Several commenters requested additional clarification regarding the notice requirements under the proposed rules, and suggested limiting such requirements to avoid unnecessary delays in the hiring process. For instance, one suggestion was for the 10 day notice requirement to only be needed once, as opposed to requiring notice for each candidate of employment on and every job posting. Another commenter recommended that posting a notice on an employer’s career website prior to a job advertisement being posted should satisfy the notice requirement. Additionally, several other commenters called for greater transparency regarding the use of AEDTs and opined that the notice requirements did not go far enough and should be expanded to inform candidates about how an AEDT will evaluate them and how the tool will maintain their data.
  • Bias Audit & Methodology Concerns: One commenter raised practical concerns for employers regarding the collection and analyses of demographic data when conducting a bias audit on an AEDT that does not capture or retain data on gender, race, or ethnicity. Seyfarth also commented that the proposed impact ratio methodology based on average scores is flawed, noting that such an analysis fails to adequately consider variability in scores and recommended further input from statistical professionals with specific experience in evaluating employment decisions. Other commenters raised additional concerns, such as the lack of guidance on how small sample sizes should be analyzed, as well as concerns regarding the intersectionality of race and gender as part of the analysis.
  • Timing of Law’s Implementation: Considering that DCWP only released LL 144’s proposed rules in late September and there is currently no public deadline for publishing the final rules implementing LL 144, several commenters suggested that DCWP delay their enforcement of the new law to allow for the rules to be finalized and to give employers sufficient time to comply with these new obligations. Commenters raised the concern that failure to postpone enforcement may lead employers, where possible, to avoid considering for employment any residents of New York City.

Now that the public record is set to close, it will be up to DCWP to consider the implications based on all of the feedback that was provided and issue their final rule. Although the timing of the final guidance is unclear, employers should be evaluating their own systems to identify what tools, if any, are in scope within NYC. Ensuring a multi-disciplinary team is in place to address the NYC requirements is key.