As employers incorporate AI into recruiting, performance management, and everyday workplace operations, compliance is becoming more complicated than simply determining whether a particular tool is legal to use. Companies also need to consider where their employees and applicants are located, what role the technology actually plays in employment decisions, how vendors handle data, and whether workers are equipped to catch mistakes made by automated systems.
Those questions are particularly pressing for employers with remote workforces. A company based in a state with relatively few AI-specific employment rules may nevertheless recruit candidates in states with more extensive requirements, potentially subjecting its hiring practices to rules it has never encountered before.
“So many employers have remote employees now,” said Dee Anna Hays, an attorney with Ogletree Deakins in Tampa, Fla. “So they’re recruiting in states where they don’t have a physical footprint,” expanding compliance requirements.
That makes the geographic scope of a job posting a compliance decision as well as a recruiting one. Before opening a remote position nationwide, employers should consider whether they actually need a national applicant pool and understand the obligations they may take on by soliciting applicants in particular jurisdictions.
Across the 16 HRX practices from recruitment and compliance to talent management and leadership development — AI is increasingly augmenting the HR function. To demonstrate how HR professionals are putting AI into practice, SHRM reached out to HR leaders to share their experiences. This installment of our AI Use Case series explores how one HR executive is using AI assistants and self-service to respond to a high volume of employee benefits questions.
This article is an example of Ethical Risk Flagging for New HR AI Use Cases. Discover 137 other AI use cases by visiting SHRM’s AI Field Manual for Employers: 138 Use Cases for AI at Work.
AI Assistance and Decision-Making
One of the more difficult questions for employers is determining when an AI tool crosses the line from assisting HR to actually influencing an employment decision. A company might view software that assigns candidates match scores as an administrative aid because a human ultimately decides whom to interview. But that distinction becomes less convincing if the score determines which resumes a recruiter sees or which applicants receive priority.
Hays pointed to employers who describe their systems as resume-screening tools that simply compare skills with job requirements. In practice, however, “if you get a high match score, you’re being fast tracked for an interview,” she said.
That practical effect could matter when determining whether laws governing automated employment decisions apply. “If I’m a plaintiff’s lawyer on the other side, I’m going to say, well, it is making a hiring decision,” she said.
The issue is even more pronounced with AI interviewers or chatbots that determine whether candidates advance. In those circumstances, she said, “it’s going to be almost impossible to avoid compliance with these state-specific requirements, because they are making a decision.”
Employers therefore should evaluate what a tool actually does within their hiring process rather than relying solely on how a vendor markets it. They should also investigate how the system was designed and tested. “I think it is very critical to vet the AI tool that you’re using,” Hays said. Employers should ask “the hard questions of what have you done to make sure that there isn’t bias built into this tool.”
Contracting with an outside provider also does not eliminate the employer’s responsibility for employment decisions. Vendor agreements may allocate costs or responsibilities between the parties, but employers remain responsible for understanding the legal requirements governing their own use of the technology. As Hays cautioned, liability “always stays with an employer,” not the vendor.
What Worked
As AI increasingly becomes part of the decision-making process in HR, its value may lie as much in supporting human judgment as in replacing manual work. David Epstein, SHRM-SCP, director of human resources and talent strategy at the nonprofit Mobilization for Justice, Inc. in New York City, said he has used AI to analyze workforce data, identify themes in employee feedback and exit interviews, and improve routine administrative tasks.
The technology, he said, can surface patterns and insights more quickly than traditional analysis. But that efficiency comes with an important caveat: “AI can save considerable time and provide useful insights, but it’s not a ‘set it and forget it’ solution,” he said. “Human review remains critical, especially in HR, where decisions can significantly impact employees.”
Epstein pointed to governance standards, bias audits, and transparency about how AI tools are used as important safeguards. Those controls can also help organizations establish who is accountable for AI-assisted decisions, what information may be fed into a system, and when an output requires additional scrutiny. With those guardrails in place, he said, AI can become “a powerful tool for better, more informed decision-making” without removing human judgment from decisions that can have significant consequences for workers.
Risks to Be Aware of in Using AI for Ethical Risk Flagging
Simply putting a person somewhere in the process may not be enough, however. Employers need to consider what that person is reviewing, how often review occurs, and whether the reviewer has enough knowledge to recognize when the technology produces a questionable result.
For a recruiting system, that could mean periodically sampling both successful and unsuccessful applicants and checking whether the AI’s assessment corresponds with the qualifications listed in the job description. A smaller employer may not have the resources for constant auditing, but even periodic checks can reveal whether a tool is systematically overlooking qualified candidates or producing results that do not make sense.
The same concern arises when AI is used to evaluate existing employees. Employers should not only review the output but consider what happens when an employee believes the system got something wrong. “It’s critical to make sure that it’s giving you good information, and also that employees have a way to object,” Hays said.
Meaningful human oversight also requires expertise and experience. A human reviewer who automatically accepts an algorithm’s recommendation may provide little protection against an inaccurate or biased result.
AI Policies Beyond Hiring
Employers also need to account for AI use that may have nothing to do with formal employment decisions. Workers can now use generative AI to draft documents, summarize information, or complete routine tasks, while AI-powered meeting tools can automatically record, transcribe, and summarize workplace conversations. Those conveniences create separate questions about confidentiality, data retention, and access.
“I think the time has come to think about whether you should have an AI policy in your [handbook] or if you keep it separate,” Hays said. Such policies can establish which tools employees may use, whether confidential information can be entered into external systems, and what types of workplace conversations should — or should not — be recorded.
Automatic meeting transcription illustrates why employers should make those decisions deliberately. Recording an ordinary strategy meeting may carry different risks from recording a harassment investigation, performance discussion, or conversation with legal counsel. It can also create large amounts of stored information that may later need to be reviewed or produced during litigation.
The basic questions, Hays said, are straightforward: “Where is that going? Who has access to that? Do we want that?”
Ultimately, employers do not need to adopt every AI capability simply because it promises efficiency. Evaluating the business need before introducing a tool can reduce both unnecessary technology and unnecessary compliance exposure. Hays’ suggested starting point is even simpler: “Do you really need this tool? Is it going to help you?”
Recommendations for HR Teams
For HR leaders looking to use AI in decision-making, a few principles can help balance innovation with risk:
- Keep humans in the loop. AI should inform consequential employment decisions, not make them unchecked. Establish clear points at which HR professionals must review and validate AI-generated recommendations.
- Know what’s behind the output. Understand what data a tool uses, what it was designed to do and where bias or inaccuracies could enter the process. Regular bias audits can help identify problems before they affect workers.
- Build governance before scaling. Set organizationwide rules for approved tools, permissible uses, data handling, documentation, and accountability rather than allowing AI practices to develop piecemeal.
- Prioritize transparency. HR teams should be able to explain when and how AI is being used and maintain documentation showing how technology contributed to significant employment decisions.
AI can make HR decisions faster and more informed — but it should not make them less human.
Was this resource helpful?
Validate your HR expertise
Earning your SHRM-CP credential makes you a recognized expert and leader in the HR field.
Related Content
Transform HR with mobile technology to streamline workflows, enhance communication, and boost employee engagement across distributed teams.
A new study found that talent development practitioners believe performance-based assessments are more effective at measuring employee learning outcomes than knowledge tests.
As AI makes deeper inroads into organizations, employee concerns about the technology continue to grow. Here's how HR leaders can address the fear of AI among their own staff, as well as in the wider organization.