In May of this year, the California Civil Rights Council (CCRC) released proposed regulations to clarify that existing laws prohibiting employment discrimination apply to automated decision-making systems that rely on artificial intelligence (AI). Employers are increasingly using these systems for recruitment and to make hiring and promotion decisions. The main provisions of the proposed regulations, which affect employers regularly employing five or more individuals, do the following:

• Require employers to maintain employment records, including automated decision-making data, for a minimum of four years;
• State that, in the context of applicants with a criminal history, automated decision-making systems alone do not replace the requirement for an individualized assessment;
• Prohibit third parties that provide services related to the administration of automated decisions systems for an employer’s use in making hiring or employment decisions from engaging in employment discrimination;
• Give examples of tests or analyses used in automated decision-making system assessments that may constitute employment discrimination based on disability, race, national origin, gender, or other protected characteristics.
• Provide definitions for terms such as “automated-decision system,” “adverse impact,” and “proxy.”

If the automated decision-making system results in an adverse impact on applicants based on protected characteristics, the system is prohibited. So, systems that analyze an applicant’s tone of voice, facial expressions, or other physical characteristics may result in an unlawful adverse impact. In addition, systems that measure a candidate’s dexterity or reaction time may adversely impact individuals with certain disabilities.

The proposed regulations provide a defense for employers in cases where there is an adverse impact or disparate treatment of a protected class. To avoid liability for discrimination, employers can show that the criteria are job-related and necessary for business and that there is no less discriminatory policy or practice available. Evidence of anti-bias testing and the response to those results can strengthen an employer’s defense.

In July, the CCRC heard public testimony regarding the proposed regulations. Employers and others argued that the scope of liability for both employers and third parties is broad and vague. Another argument put forth was that the record-keeping requirements are overly burdensome and implicate privacy concerns. For these reasons, the parties asserted that the proposed regulations could result in a chilling effect on employer use of AI. If the proposed regulations are approved, they will be subject to final review by an administrative law judge.

The proposed regulations are related to Assembly Bill 2930, which is currently before the California legislature. The bill would require employers, among others, who use automated decision-making systems to make “consequential decisions,” such as employment-related decisions, to notify the applicant that an AI decision tool is being used. Moreover, the bill would prohibit AI tools from being used if there is a risk of algorithmic discrimination.

Lastly, a case recently decided under federal law in the U.S. District Court for the Northern District of California, found that a vendor of AI hiring tools, Workday, can be liable for employment discrimination under the “agent” theory of liability as a result of an employer using such vendor’s tools. The court found it significant that Workday’s tools “are alleged to perform a traditional hiring function of rejecting candidates at the screening stage and recommending who to advance to subsequent stages, through the use of artificial intelligence and machine learning.”

With the ever-increasing use of AI as a tool for employers to make employment decisions, there will undoubtedly be an increase in challenges to their use. Employers using such tools should, in addition to ensuring that hiring criteria are job related and necessary for business, regularly conduct anti-bias testing and respond proactively to the results.