Following its announcement delaying the enforcement of Local Law 2021/144 (LL144) to April 15, 2023, the DCWP released a second set of proposed regulations implementing LL144 (Updated Proposal). The Proposed Regulations address a number of ambiguities in the underlying statute and include several notable changes from the previous set of proposed regulations (Initial Proposal). We describe these changes below.
More Onerous “Independent Auditor” Requirements
The Updated Proposal has made it more challenging for a person to quality as an “independent auditor.” The Initial Proposal defined “independent auditor” as “a person or group that is not involved in using or developing an AEDT.” This definition would have provided greater flexibility for organizations to audit AEDTs by leveraging appropriate internal resources, such as internal audit functions or data scientists on teams that do not use or develop the AEDT.
The Updated Proposal introduces new requirements and expressly disqualifies several categories of persons. The updated definition of “independent auditor” requires that the person be “capable of exercising objective and impartial judgment on all issues within the scope of a bias audit of an AEDT.” In addition, the a person or group would be expressly disqualified from being an independent auditor if the person or group:
is or was involved in using, developing, or distributing the AEDT;
at any point in the audit, had an employment relationship with an entity seeking to use the AEDT or with an entity that developed or distributed the AEDT; or
at any point in the audit, had a financial interest or indirect material interest in an employer using, or vendor developing or distributing, the AEDT.
If finalized, these additional requirements may present new challenges to organizations. Organizations may find it difficult to identify sufficiently independent experts who have appropriate technical expertise and understanding of the particular context in which the AEDT is being used. Organizations may also need to develop a process through which to evaluate whether potential auditors are sufficiently independent from the organization to provide a valid audit.
Bias Audits Frequency and Standards
The Updated Proposal affects several aspects relating to bias audits.
First, the Updated Proposal would prohibit employers and employment agencies from using AEDTs if more than one year has passed since the most recent bias audit. The text in LL144 does not clearly require bias audits to be conducted at least annually; however, the DCWP has now advanced an affirmative interpretation of the text by introducing this requirement.
Second, the Updated Proposal would regulate the placement and content of bias audit summaries. The Updated Proposal requires that bias audit results be published in a “clear and conspicuous manner” on the employment section of an employer’s website and must include the source and explanation of the data used to conduct the bias audit, the date of the most recent audit, the date the AEDT was distributed, and “the selection rates and impact ratios” for the categories considered in the audit. The published summaries must also be posted for at least six months after last using the AEDT for an employment decision.
Third, the Updated Proposal would require bias audits to separately calculate the selection rate and impact ratio for sex categories, race and ethnicity categories, and intersectional categories. These calculations also extend to an AEDT used for group classifications.
New Data Requirements
The Updated Proposal introduces new concepts of “historical data” and “test data” and regulates the situations in which either type of data may be used in a bias audit. “Historical Data,” which means “data collected during an employer or employment agency’s use of an AEDT to assess candidates for employment or employees for promotion,” is the primary source of data for bias audits and must be used if it is available.
In the event “historical data” is not available to conduct a “statistically significant bias audit,” the Updated Proposal now allows for “test data” to be used. “Test data” is defined in the negative: it is any data used for a bias audit that is not “historical data.” In the event “test data” is used, the bias audit summary must explain why “historical data” was not used and describe how the “test data” was “generated and obtained.”
The next steps for entities who may be affected by LL144 are as follows:
Consider whether to provide comments on the Updated Proposal by January 23, 2023. LL144 is a first-of-its-kind law in the United States and the outcome of the rulemaking process may influence future legislative developments across the nation. Indeed, New Jersey has already introduced a state-wide bill similar to LL144, and the California Privacy Protection Agency will soon be undertaking a rulemaking on automated decision-making that may also apply in the human resources context.
Prepare for enforcement under LL144 beginning April 15, 2023. Such preparation may involve several activities such as:
Evaluating whether any persons who conducted a bias audit for the affected entity adequately meet the “independent auditor” standard established in the Updated Regulations. Employers and employment agencies who intended to rely on the bias audit results from AEDT developers may wish to similarly examine whether the auditors were sufficiently independent and consider the extent to which the audit was based on historical data or test data.
Updating public disclosures, including bias audit summaries, to reflect the Updated Regulations’ placement and content requirements.
Watch the legislative landscape across the U.S. As we noted above, LL144 may inspire novel legislation or regulations across the nation. Companies that use automated tools to support employment decisions may wish to keep an eye for developments in this space, even if they are not directly subject to LL144.
Authored by Kenneth Kirschner, W. James Denvil, Filippo A. Raso, and Zach Siegel. Muhammad Burney, a law clerk in our New York office, contributed to this post.