AI in employment decisions: Balancing innovation, fairness and legal risk

Artificial intelligence (AI) is here to stay and will only become more ingrained in our lives as technology evolves. We all (hopefully) will be able to control how pervasive it will be in our personal lives, but there is little doubt about the fast-paced progression of AI in business.

Despite the fast-paced progression, AI is still in its infancy, and the unknowns outweigh the knowns. The value gained from AI, even if at a profound level, still produces some unintended, negative consequences. Society, regulators and prospective plaintiffs, will and are monitoring these risks closely, especially those negative consequences that impact people the most — namely risks involving physical security, privacy, access to information, reputation and discrimination (including employment discrimination).

AI is being used throughout the employment life cycle — from recruiting to termination decisions and everything in between. Below, we will take a closer look at the various uses within the employment life cycle and how they may lead to EPL exposure.

Recruiting

Some common examples of AI use in recruiting include using an AI tool to search for potential candidates, screen resumes, rank candidates based on algorithms aligned with the organization’s needs, schedule interviews, summarize interviews and AI assisted video interview analysis to evaluate candidates. While some of the uses are benign, others may create an additional avenue for EPL exposure.

Two notable class action lawsuits are pending regarding the use of AI hiring tools to screen resumes. One is against the employer, and the other case is against the AI hiring tool vendor. Both allege violations of Title VII, state anti-discrimination laws and disparate impact.

The case against the vendor is significant and being closely watched in that it highlights potential vendor liability for employment discrimination claims. In a recent motion, the Court ruled  that the allegations that the AI vendor is an agent of its client-employers and therefore falls within the definition of “employer” under federal anti-discrimination laws may proceed to the discovery stage. This means that AI vendors may have direct liability for employment discrimination claims under federal anti-discrimination laws, significantly expanding their scope of liability.

In addition, the lawsuits allege a disparate impact theory of liability. While Executive Order 14281 directs federal agencies to eliminate disparate impact enforcement, it does not affect private litigation or state law claims so those theories may still proceed in any given case.

Termination

Employers are also using AI to assist with layoffs, reduction in force and termination decisions. Many of the AI tools used rely on performance and productivity metrics. Some examine an employee’s attendance, which includes a look at their paid time off, sick days and medical leave absences. Other tools look at salary and tenure as factors as well.

This does not come without risk. Potential EPL exposures include discrimination and wrongful termination claims. For example, a lawsuit was recently filed against a large tech company by current and former employees wherein they allege that the company used AI to make their recent layoff decisions in a way that was discriminatory. Specifically, the claimants allege that the AI tools utilized flawed metrics in that it did not consider if an employee had taken or was on leave, which would naturally impact their level of productivity versus someone who had not taken leave. They allege violations of state and federal anti-discrimination laws, specifically ones protecting disabled individuals, pregnant individuals and those taking medical leave

Other potential EPL exposures

In addition to recruiting and termination decisions, AI is also being used to provide recommendations for compensation structures and promotions, to monitor employee behavior and productivity and more. Below are some of the potential risks associated with some of the ways in which AI is being used:

• Failure to promote: AI assisted decisions for promotions and compensation structures
• Privacy: AI tools incorporating private data outside of the intended and/or disclosed purpose (e.g., biometric data); use of AI to monitor employee behavior and productivity in the workplace (invasion of privacy issues)
• Harassment: Employees using AI to create fake, inappropriate, harassing content against another employee and/or third party

Legal landscape

There is currently no federal law in place specifically regulating the use of AI in the employment context. There is, however, a growing patchwork of state laws that are widely inconsistent and differ on their approaches to notice and audit obligations and enforcement. These patchwork of laws create significant challenges for multi-state employers.

For example, in Connecticut, the Governor will be signing into law, Senate Bill 5 (Artificial Intelligence Responsibility and Transparency Act) that will impose new obligations on developers of AI and deployers (i.e., employers). The Act further clarifies by amending the Connecticut Fair Employment Practices Act, that “AI is not a defense” to a discrimination claim. While an employer’s bias audits and testing may be presented as mitigating evidence, an employer cannot escape liability by stating they relied on the vendor’s algorithm. The Act also provides that if an employer is filing a disclosure under the Connecticut WARN Act, they must disclose whether AI was used in making any layoff decisions.

The Texas Responsible Artificial Intelligence Governance Act, on the other hand, only prohibits intentional discrimination and does not have any audit or notice obligations. Whereas New York City and others require employers to conduct regular audits and provide notice of their use of AI. Further, in Illinois, HB 3773 specifically prohibits discrimination based on certain protected characteristics and includes unintentional, disparate impact discrimination.

In addition to this growing patchwork of laws, Title VII, the Age Discrimination in Employment Act, the Americans with Disabilities Act and state anti-discrimination laws remain in full force. While the state statutes being passed do not include private rights of action, pending litigation related to the use of AI in employment decisions cite legal theories under traditional anti-discrimination laws.

Conclusion

There is no doubt that AI is a necessary tool in today’s workplace. While it does enhance the way we work and eliminates some inefficiencies, it is not a tool of substance. AI does not eliminate employer accountability and risk. Human oversight and input are imperative.

When implementing AI in HR functions, robust governance, testing, transparency and human review are essential to reducing EPL exposure. The need for efficiency must be balanced against the need to ensure decisions are fair, can be validated and explained and comply with existing employment laws.