AI in HR: Legal Risks and Why Employers Must Audit Their Hiring Tools

This article is part of an ongoing legal series designed to provide insight and practical guidance on current and emerging workplace compliance issues. These insights shared by lawyers are based on their interpretation of existing regulations and proposed changes, and intended for informational purposes, not to be regarded as legal advice. 

Using AI in HR can speed up processes, improve efficiency, and help organizations get and stay organized. Yet AI poses risks, especially if organizations do not fully understand the system or are not careful when using AI tools for HR functions. If an AI tool causes an employer to make discriminatory decisions, the employer and the vendor could be held responsible.

By auditing hiring tools, organizations can better understand how these systems operate, identify potential blind spots, and confirm that the tools function as intended without introducing impermissible bias.

What AI Tools for HR Are Available?

With each year, more HR AI tools become available. Organizations may use tools for:

• Recruiting—using an Applicant Tracking System (ATS) to compare and score applicants, video tools that evaluate interviews, and chatbots that interact with applicants;
• Training and development—interfaces and platforms to guide employees through training materials and track their progress;
• Scheduling—systems that schedule and track meetings, interviews, and similar work events; and
• Performance management and engagement—generating performance reviews, using tools to predict compensation, analyzing productivity metrics, and taking notes for meetings.

Many of these tools can evaluate, summarize, and draw conclusions from the data you provide. Generally, AI tools rely on complicated programming and large language models (LLMs). These tools are models that predict and generate text based on associations within words in the data the tools train on.

What Are the AI Legal Risks for HR Teams and Professionals?

Many of the AI legal risks for HR center on the possibility that AI tools may make errors or reach discriminatory conclusions. A business may be liable for relying on tools that cause the most qualified applicant not to be interviewed or hired for a position or is making decisions based on protected characteristics, though the implications vary depending on what goes wrong and why.

AI Tool Errors

One of the most frustrating aspects of AI tools is the possibility of hallucination. This occurs when the model generates partially or fully fabricated information. In these cases, the output may not reflect the data provided but instead reflects the system’s attempt to predict a plausible response.

Sometimes AI tools also generate errors when they receive excessive or poorly structured input, as the system tries to determine which information is most relevant. Additionally, tools sometimes have processing errors and may simply fail to perform tasks.

Anti-Discrimination Laws

If not used carefully, AI in HR may create risk under state, federal, and local anti-discrimination laws, including:

• Title VII of the Civil Rights Act of 1964—prohibits discrimination in the terms or conditions of employment based on race, color, sex, national origin, and religion;
• The Americans with Disabilities Act of 1990 (ADA)—prohibits discrimination against individuals with disabilities and requires reasonable accommodations for individuals qualified and able to perform the work;
• The Age Discrimination in Employment Act of 1967 (ADEA)—prohibits discrimination based on age, specifically discrimination against job applicants age 40 or older; and
• The Equal Pay Act—requires that employers provide equal pay for substantially equal work, regardless of an employee’s gender or sex.

Several states have also passed or are considering laws that regulate the use of AI in hiring practices. However, the current presidential administration has issued an executive order aimed at curbing state efforts to regulate AI, raising questions regarding the enforceability of such laws.

Given what is understood about implicit bias, there is a genuine risk that a large language model that generates expected outputs may produce a biased, discriminatory conclusion that violates one or more of these laws.

Implicit Bias and Algorithmic Discrimination

Consider an example of implicit bias. In a study, researchers sent identical resumes to various locations, changing only the candidate's name. After receiving callbacks for interviews, researchers concluded that white-sounding names, especially male white-sounding names, resulted in significantly more attention than non-white-sounding names. An AI tool trained on data with those same biases may replicate them, posing a real risk. Some call that risk algorithmic discrimination, and it bears striking similarities to implicit bias.

This risk makes AI bias audits an imperative practice to implement into your processes. If not audited routinely, you could be setting yourself up for risk as employers are liable for the tools they use. However, given a recent Colorado AI law, the responsibility could also shift back to the vendor if the employer uses the tool the way it was intended to be used.

AI facial recognition and video scanning tools have also proved discriminatory in the criminal justice field. Facial recognition technology may misidentify up to 35% of black women. HR tools that use facial scanning likely rely on the same underlying software, which, unfortunately, may be riddled with bias.

Court cases already exist challenging the use of AI tools in employment decisions. For example, individuals filed a class action lawsuit against the online platform Workday, alleging age discrimination. Others are suing Sirius XM for allegedly using AI tools that engaged in racial discrimination.

AI Tools and Privacy

AI tools also collect large amounts of data. Protecting the data of all applicants and employees is critical, as organizations may be responsible for damages applicants incur due to data breaches. Investing time in understanding the specific data privacy implications of the tools used can save time and money in the long run, especially if the tools include a database hosted remotely, where an organization may have less control over the data.

If operating in the European Union, organizations must also comply with the EU Artificial Intelligence Act, which regulates the use of AI tools for human resources. The law defines most AI HR tools as high risk and requires organizations that use them to meet special requirements.

What Are the Practical Risks of Using AI in HR?

There are also practical risks of using AI in HR; one of the most common is misunderstanding what an LLM actually does. An AI tool is a complicated program that generates outputs based on patterns in data and is not inherently accurate or objective. The systems are not infallible, and responsible use of AI tools in HR requires awareness of their drawbacks and risks.

Another practical, related risk is overconfidence in the tool. Organizations should regularly compare AI-generated outputs with human judgment to identify inconsistencies or errors.

Auditing Your Hiring Tools

If incorporating AI in HR, organizations should regularly assess how the tool is performing. This includes reviewing outcomes, testing for potential bias, and confirming that the system aligns with internal policies and applicable law. To get the most out of AI tools in HR, begin by understanding their practical uses, risks, and limitations.

Outsolve supports organizations as they evaluate AI tools, review HR processes, and align workplace practices with evolving legal requirements.

To learn more about how to balance AI, HR, and the law, reach out to Outsolve.