A growing number of employers are turning to artificial intelligence (“AI”) tools to assist in recruiting and other employment decisions. According to Forbes, almost all Fortune 500 companies use talent-sifting software, and more than half of human resource leaders in the U.S. leverage predictive algorithms to support hiring. Widespread adoption of these tools has led to concerns from regulators and legislators that they may be inadvertently discriminating, for example, by:

  • Penalizing job candidates with gaps in their resumes, leading to a bias against older women who have taken time off for childcare;
  • Recommending candidates for interviews who resemble the company’s current leadership, which is not diverse; or
  • Using automated games that are unfairly difficult for individuals with disabilities to evaluate employees for promotions, even though they could do the job with a reasonable accommodation.

New York City is one of the first jurisdictions to pass a law aimed at reducing bias in automated employment decisions, which becomes effective on January 1, 2023. The Automated Employment Decision Tool Law (“AEDT”) places compliance obligations on employers in New York City that use AI tools, rather than software vendors who create the tools. Similar laws are likely to be enacted in other jurisdictions. Accordingly, companies should pay close attention to any AI tools or algorithms being used to manage human capital to ensure that they are compliant with these emerging requirements.

In this Data Blog post, we discuss the key requirements of the new City law, the growing scrutiny of AI-based hiring tools in other jurisdictions, and practical steps companies can consider taking to reduce their legal and regulatory risks related to their use of these automated tools.

What Does the AEDT Require?

The AEDT applies to companies located in New York City, that are hiring or promoting City residents, for jobs that are located in the City using “automated employment decision tools” to “replace” or “substantially assist” decision-making in hiring or promotions. What is unclear is whether it applies in other circumstances (e.g., companies located outside of the City hiring New York residents, companies in the City hiring applicants from outside the city, etc.). Hopefully the full scope of the AEDT will be made clearer in the forthcoming rulemaking process.

The term “automated employment decision tools” is broadly defined as any “computational process, derived from machine learning, statistical modeling, data analytics, or artificial intelligence” that “issues a simplified output.” Given the breadth of this definition, a wide variety of automated tools will likely be covered by this law, even if they do not employ true AI, including many game-based tests and some resume review tools and automated personality assessments. Additionally, the decision need not be fully automated for the AEDT to apply. Any automated tool that “substantially assists” a human in reaching their decision (for example, by evaluating or recommending candidates) may fall within the scope of the law.

For companies subject to the AEDT, compliance obligations include:

  • Conducting an Independent, Annual Bias Audit. Companies must ensure that their automated employment decision tools are subjected to a “bias audit,” conducted no more than one year prior to the use of the tool. It appears that this must be done annually. Although the law provides very little information on the substantive requirements of this bias audit, it does define the audit as “an impartial evaluation by an independent auditor” that includes testing the tool to assess its “disparate impact” on persons based on gender, race, or ethnicity. The term “disparate impact” is not defined, nor does the law specify a methodology for conducting the bias audit, or how companies should assess bias if they do not collect data on the race, ethnicity or gender of their job applicants. One possible approach would be to use the “four-fifths rule,” as defined in federal anti-discrimination regulations on employee selection procedures promulgated by the Equal Employment Opportunity Commission (“EEOC”), whereby a selection rate of a protected group that is less than 80% of the rate for the group with the highest rate constitutes evidence of “disparate impact.” The AEDT also does not specify the level of “independence” required from the auditor (e.g., whether the internal audit function is sufficiently independent and whether the auditor must do more than review testing performed by the company or the software provider).
  • Providing Disclosures. Before using a covered tool, companies must publish a “summary” of the results of the bias audit on its website, along with the distribution date of the tool. Additionally, the company must publish on its website, or provide to a candidate or employee within 30 days of their request, “the type of data collected for the automated employment decision tool, the source of such data and the employer or employment agency’s data retention policy.”
  • Notifying Candidates or Employees. At least 10 days before using a covered tool, a company must provide notices to candidates and employees residing in New York City: (i) that an automated employment decision tool will be used to assess their employment or candidacy; (ii) the job qualifications and characteristics the tool will assessing; and (iii) that the candidate may request an “alternative selection procedure” or “accommodation.”
  • Providing an Accommodation or Alternative Selection Process. The AEDT requires that companies provide candidates or employees residing in New York City with the ability to request an alternative selection process or accommodation; it does not, however, state what these alternatives or accommodations should entail.

Companies that are found to not be in compliance will face penalties of $375 for a first violation, $1,350 for a second violation, and $1,500 for a third violation and any subsequent violations. Each day that a company uses a covered tool in noncompliance with the law constitutes a separate violation, as does the failure to provide any required notice. New York City’s Corporation Counsel may bring proceedings to enforce the AEDT. While the AEDT does not include a private right of action, it does not preclude private plaintiffs from bringing civil actions related to a company’s practices or automated tools (including, for example, discrimination claims). The New York City Commission on Human Rights may also enforce the law.

On June 6, 2022, New York City’s Department of Consumer and Worker Protection conducted a hearing pertaining to the AEDT, and a further rule-making process is currently anticipated. At the time of writing, however, it remains unclear when this rule-making will occur or what topics it may cover.

Other Emerging AI Hiring Laws and Regulations

Outside of New York City, several states currently have enacted or proposed various laws or regulations that apply to AI tools used for hiring or promotions. For example:

  • Washington, D.C. In 2021, Washington, D.C. introduced legislation that would prohibit certain companies from making algorithmic decisions about “important life opportunities”—including employment offers—on the basis of actual or perceived protected classes. This law would require companies to audit their algorithmic determination practices on an annual basis for potential disparate impact and report this information to the Office of the Attorney General, as well as to preserve an audit trail for five years. Additionally, companies using vendor-provided models would need to obtain written agreements that the vendor has implemented and maintained measures “reasonably designed to ensure” that the company complies with this law. The bill also contains a proposed private right of action and is currently under D.C. Council review with a public hearing scheduled for September 2022.
  • On March 15, 2022, the California Fair Employment and Housing Council published draft changes to their employment and discrimination laws, which, if passed, would impose liability on companies or third-party agencies administering artificial intelligence tools that “screen out or tend to screen out an applicant or class of employees on the basis” of a protected characteristic and create a private right of action for those who are discriminated against by the AI tools. The regulations are currently pending and will be subject to a public comment period before taking effect.
  • In 2019, Illinois passed its Artificial Intelligence Video Interview Act, which gives job applicants the right to know and provide consent, before their interview, that (a) that AI may be used to analyze the video, and (b) what characteristics will be analyzed. The employer is restricted in sharing the applicant’s video and must also destroy it within 30 days of the applicant’s request. A recent amendment, effective January 1, 2022, has also required employers that rely solely on AI video analysis for determining who to interview in person to provide annual reports of demographic data to the state’s Department of Commerce and Economic Opportunity, including “the race and ethnicity of applicants who are and are not afforded the opportunity for an in-person interview after the use of artificial intelligence analysis; and . . . the race and ethnicity of applicants who are hired.”
  • In 2020, Maryland passed a law prohibiting employers from using facial recognition technology during pre-employment job interviews (including in the context of AI tools) without the applicant’s written consent.

Whether or not covered by these specific AEDT Laws, most employers using AI are subject to other general anti-discrimination laws, including Title VII of the Civil Rights Act. Given the rise of AI-related hiring tools, the EEOC has stated that it remains focused on ensuring that AI does not “become a high-tech pathway to discrimination.” Most recently, in May 2022, the EEOC issued its first non-binding technical guidance regarding how employers’ use of AI may violate existing requirements under the Americans with Disabilities Act (“ADA”). Among other things, the EEOC recommends that employers give applicants or employees notice that they are undergoing an assessment by an AI tool, which traits or characteristics the tool is designed to measure, and that they may request a reasonable accommodation or exemption from the tool. The Department of Justice also joined the EEOC in warning of potential risk that the use of AI tools by employers may “result in unlawful discrimination against certain groups of applicants, including people with disabilities.”

Outside of the United States, AI tools used for hiring and recruiting have drawn scrutiny from European lawmakers. As we have previously discussed, the European Commission’s draft AI Act would place potentially onerous regulatory and disclosure obligations on any AI systems classified as “high risk,” such as AI systems that are used for recruiting and workplace management, including evaluating candidates through interviews, making decisions concerning promotions or termination, or monitoring and evaluating employee performance or behavior. Although the AI Act is still in being refined through the European Union’s legislative process, it is likely that at least some hiring and promotion systems will be classified as “high risk” in the final version when passed.

Four Tips for Complying with AEDT Laws

In light of these emerging requirements, employers using AI tools to hire or promote talent should consider the following measures to reduce their legal and regulatory risks:

  1. Identify Which Models, Algorithms, or Other Tools Are Subject to AEDT Laws. Companies should determine whether their employment tools are subject to AEDT Laws because they (a) “replace” or “substantially assist” human decision-making and (b) involve a simplified output from a computational process, including AI, machine learning, data analytics or statistical tools. Given the extremely broad definition, many sophisticated hiring tools are likely to qualify.
  2. Consider Whether to Leverage the Vendor’s Bias Audit(s) of the Tool. Although the AEDT itself squarely places the burden of compliance on employers using AI hiring and recruiting tools, the vendor providing those tools may be best positioned to conduct an audit to assess the tools for potential disparate impact. Companies will need to determine on a case-by-case basis whether they should rely on the vendor’s bias audit, but at a minimum, such an audit must have been conducted (a) by an independent auditor, (b) no more than 12 months prior to the company’s use of the tool, and (c) involve an evaluation of potential disparate impact risks. Companies may also want to consider whether the vendor’s bias audit is applicable to the way it uses the tool for its employment decisions.
  3. Determine What Other Testing and Evaluation Steps Should Be Included in the Bias Audit. In addition to the assessment of hiring tools for disparate impact, companies might consider whether the bias audit should include a qualitative assessment of the relevant policies and practices by the company or the vendor of the AI tool, including:
  • Whether the company and/or the vendor have a written policy on responsible AI use that applies to tools used for hiring or promotions;
  • Whether relevant individuals at the company or vendor (such as those who will be using the tool or who designed the tool) have received training on detecting and preventing bias in the use of AI hiring and promotion tools; and
  • Whether, in order to reduce the risk of bias, the vendor has identified criteria that should not be used when operating the tool, (e.g., name, race, ethnicity, sex or gender, sexual orientation, gender identity or expression, age, religion, national origin, disability status, family or marital status, genetic characteristics, information regarding a conviction for which a pardon has been granted or a record suspended, or protected veteran or other uniformed status), as well as proxies for these characteristics (e.g., address, zip code, etc.).
  1. Evaluate What Sort of Accommodation or Alternative Selection Processes Should Be Given to Candidates or Employees. The AEDT requires employers to provide notice to candidates or employees that they may request an accommodation or alternative selection process. The AEDT is silent, however, as to what kinds of accommodations and alternative selection processes should be provided, and in what circumstances. Companies should nevertheless consider practical means of providing human-based selection systems to candidates or employees that elect to opt-out of the automated tool, especially for persons who may have a disability. For example, companies using game-based assessment tools with complex graphics may want to offer candidates or employees with visual impairments an alternative screening process. Indeed, any accommodations offered by employers should be evaluated for compliance with the ADA and other applicable laws.

To subscribe to our Data Blog, please click here.

Author

Avi Gesser is Co-Chair of the Debevoise Data Strategy & Security Group. His practice focuses on advising major companies on a wide range of cybersecurity, privacy and artificial intelligence matters. He can be reached at agesser@debevoise.com.

Author

Jyotin Hamid, a partner in the New York office, is a seasoned litigator with extensive courtroom experience. He handles a diverse array of complex commercial litigation matters, with a particular focus on employment litigation and intellectual property disputes. He has represented major companies in their most challenging commercial matters. In the employment area, he has successfully handled numerous whistleblower, discrimination, contract, compensation and corporate raiding litigations involving high-level executives in a broad range of industries. Mr. Hamid also counsels employers on their most sensitive personnel matters, including investigations of alleged executive misconduct. He is also deeply involved in Debevoise’s market-leading intellectual property practice, and he has litigated trademark and trade dress cases involving some of the most well-known brands in the world.

Author

Tricia Bozyk Sherno is a member of Debevoise's Litigation Department, concentrating in employment and general commercial litigation. She has a broad-gauged employment law practice, with experience representing clients in matters involving discrimination and harassment, contracts, corporate raiding and compensation across a broad range of industries. She can be reached at tbsherno@debevoise.com.

Author

Anna R. Gressel is a senior associate and a member of the firm’s Data Strategy and Security Group and its FinTech and Technology practices. Her practice focuses on representing clients in regulatory investigations, supervisory examinations, and civil litigation related to artificial intelligence (“AI”) and other emerging technologies. Ms. Gressel has a deep knowledge of regulations, supervisory expectations, and industry best practices with respect to AI governance and compliance. She regularly advises boards and senior legal executives on governance, risk, and liability issues relating to AI, privacy, and data governance. She can be reached at argressel@debevoise.com.

Author

Scott M. Caravello is an associate in the litigation department. He can be reached at smcaravello@debevoise.com

Author

Rachel Tennell is an associate in the litigation department. She can be reached at rmtennell@debevoise.com