US Department of Labor Publishes Guidelines Addressing Use of AI in Employment Decisions

The US Department of Labor’s Office of Federal Contract Compliance Programs (OFCCP) recently issued guidance for the use of artificial intelligence (AI) by federal contractors in employment decisions.

Building on President Joseph Biden’s October 30, 2023 Executive Order 14110 (Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence), the guidelines acknowledge both the promises and risks of applying AI in hiring, show an intent to treat and investigate AI models in a similar manner to other selection procedures, and emphasize that the burden is on employers to maintain adequate records and answer questions involving AI models.

In a potential nudge to large-scale industries, such as defense and utilities, the OFCCP presents an example of using AI as an initial filter when filling positions that draw thousands of applicants.

The guidance recognizes AI’s usefulness but also stresses the need for guardrails and proactive monitoring, noting that “AI has the potential to embed bias and discrimination into a range of employment decision-making processes.”

The guidance is broken into two major sections: (1) “Common Questions About the Use of AI and EEO” and (2) “Promising Practices for the Development and Use of Artificial Intelligence in the EEO Context.” Together, these sections provide employers with a roadmap for compliance with the agency’s interpretation of contractor obligations relating to affirmative action and the prevention of unlawful discrimination. The guidance also reinforces the Equal Employment Opportunity Commission’s existing position that equal employment opportunity obligations apply fully to any use of AI in employment decision-making.

Although the guidelines only apply to federal contractors, private employers should take note of the ways in which AI can implicate anti-discrimination law.

KEY TAKEAWAYS

The OFCCP provides insights on how the agency will treat AI models when evaluating federal contractor compliance, such as during audits. The OFCCP plans to treat AI models like other selection procedures, so federal contractors need to be able to articulate how their AI models operate and maintain records that substantiate these explanations.

Likewise, federal contractors also need to provide adequate justification for any adverse impacts on protected groups that result from the application of AI. Federal contractors cannot shield themselves from liability through the use of third-party tools or delegate these obligations to vendors. Consistent with the “Promising Practices” noted in the guidance, employers can design and proactively monitor their use of AI systems to comply with the agency’s interpretation of their obligations.

EEO Obligations Apply to Federal Contractors’ Use of AI in Employment Decisions

The OFCCP states that federal contractors must comply with all equal employment opportunity (EEO) obligations in their use of AI systems for employment decisions.

The guidance takes an expansive view of AI, which it defines as a “machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments.” Likewise, the OFCCP’s definition of “employment decisions” is broad and covers “all aspects of the terms and conditions of employment” including hiring or termination, pay or promotion decisions, workplace training programs, and workplace surveillance or management systems.

The guidance also states that federal contractors must

• maintain records on AI models and ensure confidentiality consistent with OFCCP regulatory requirements;
• cooperate with the OFCCP by providing requested information on their AI systems; and
• make available reasonable accommodations for physical or mental limitations of otherwise qualified applicants or employees with a disability, unless doing so would impose an undue hardship on the contractor.

Additionally, if a selection procedure that relies on AI has an adverse impact on members of any race, sex, or ethnic group, employers must validate the system consistent with nondiscrimination laws and the Uniform Guidelines on Employee Selection Procedures. This includes articulating the business needs for the AI model, analyzing job-relatedness, engaging in bias audits and independent assessments, and exploring less discriminatory, alternative selection procedures.

Obligations and Liability Remain on the Employer

The guidance also emphasizes that federal contractors cannot shield themselves from EEO obligations and potential liability with the use of third-party products. Even where an employer obtains an AI system from a third-party vendor, the employer remains responsible for meeting its nondiscrimination and affirmative action obligations under the laws enforced by the OFCCP. These laws do not impose separate obligations on vendors so federal contractors cannot escape liability by deferring to a third party.

Relatedly, federal contractor employers should ensure that any third-party AI models they use are transparent, substantiated with adequate documentation, and subject to regular independent bias audits.

Federal Contractors Should Proactively Mitigate Risks Associated with the Use of AI

Finally, the OFCCP’s “Promising Practices” for the use of AI provides a framework for federal contractors to consider when designing or utilizing an AI system in making employment decisions. The Promising Practices address four core areas: (1) notice to applicants; (2) using an AI system; (3) obtaining a vendor-created AI system; and (4) ensuring adequate accessibility and disability inclusion.

Before adopting an AI system, federal contractors should provide advance notice to applicants or employees that it intends to use AI, provide information on how employees or applicants can request an accommodation, explain what data will be captured and whether it can be reviewed/corrected, and safeguard the privacy of jobseekers and employees.

Once the AI model goes live, the employer should engage with applicants/employees, follow standard processes for accommodation requests, routinely monitor any adverse impacts, verify that documentation is easily accessible, and ensure that there is sufficient human oversight from trained individuals and broader oversight from governance structures.

The guidance also provides additional recommendations when contracting with a third-party vendor for an AI system. Employers should ensure that vendor contracts include specific provisions on maintaining easily accessible records and strong data privacy guardrails consistent with OFCCP regulatory requirements.

Federal contractors should also confirm that, beyond making accommodations available, the AI system is accessible and incorporates inclusive design elements. Employers should perform ongoing testing and monitoring of the accessibility for the system.

CONCLUSION

Federal contractor employers now have a roadmap for OFCCP compliance in the AI space. While other developments may provide additional details for this roadmap, at a minimum, federal contractor employers should understand that equal employment opportunity obligations apply fully to any use of AI in employment decision-making. Private employers should also take note of the ways in which AI can implicate anti-discrimination law.