Employers are increasingly using artificial intelligence and other algorithmic tools to support workplace decisions, including recruiting, screening, interviewing, promotion, workforce planning, and performance management. These tools can improve efficiency and consistency, but they also introduce important compliance, reputational, and employee-relations considerations. Two concepts that often arise in AI governance are bias audits and validation testing. Although related, they serve different purposes.

A bias audit generally evaluates whether the use of a tool is associated with materially different outcomes across protected or demographic groups. Depending on the jurisdiction and the tool at issue, a bias audit may be legally required before use. For example, New York City’s automated employment decision tool law requires certain employers and employment agencies to obtain a bias audit within one year before using covered tools and to provide related notices and disclosures. And other states have laws that affect the need for bias testing or are considering such requirements. Even where no specific audit law applies, employers may decide that an audit is appropriate as part of responsible AI governance, particularly when a tool affects access to job opportunities or advancement.

Validation testing, by contrast, focuses on whether a selection procedure is job-related and appropriately measures what it is intended to measure. This concept is not new. The Uniform Guidelines on Employee Selection Procedures provide a long-standing framework for evaluating employment tests and other selection procedures used in employment decisions in compliance with Title VII of the Civil Rights Act. In the AI context, a validation study may be especially important where an algorithm ranks, scores, recommends, or screens individuals based on characteristics that are intended to predict job success. If an employer’s use of an AI tool is challenged, a validation study may be critically important to the employer’s defense.

As a general governance matter, employers should consider bias audits and/or validation testing before deploying an AI tool that materially influences employment decisions. They should also consider reassessment when the tool is modified, used for a new role or population, applied in a new jurisdiction, or when the employer’s own workforce or applicant data materially changes. Ongoing monitoring is critical because a tool that performs acceptably at one point in time may produce different results as business practices, labor markets, or applicant pools evolve.

The need for testing should be assessed early in the procurement process. Employers should ask vendors what the tool does, what data it uses, whether it has been audited or validated, what assumptions underlie the model, and what documentation is available. Employers should also avoid assuming that a vendor’s general statements about fairness, accuracy, or compliance are sufficient for the employer’s particular use case.

A practical AI governance program typically includes an inventory of AI-enabled employment tools, a risk-based review before deployment, appropriate legal and human resources oversight, documentation of decision-making, and periodic reevaluation.

If you have questions about AI governance for your business, contact a Jackson Lewis attorney to discuss.