In an effort to spark more artificial intelligence use in federal hiring, the Office of Personnel Management is advising agencies that several common HR use cases generally aren’t considered “high impact” for the purposes of a federal policy governing the technology.
The Thursday memo to department heads from OPM Director Scott Kupor argues that failure to adopt AI in federal hiring may be compromising “efficiency and quality” of that process. It shares OPM’s own assessment of how federal AI governance policy applies to the human capital space to provide agencies with a roadmap for use.
Specifically, OPM concludes AI used for creating job announcements, evaluating applicants, reviewing files before an offer, and evaluation of hiring metrics isn’t considered high-impact in most instances. That means those uses wouldn’t be required to follow corresponding minimum risk management practices, such as conducting pre-deployment testing, conducting an impact assessment, and monitoring the tool for potential adverse impacts.
In an accompanying blog post, OPM Chief Information Officer Adam Starr explained the memo is an attempt to aid “overly cautious” agencies that felt AI applications in hiring are considered high impact.
“Too often, agencies were defaulting to the view that any use of AI in hiring was subject to the high-impact standard. In reality, that is not the case,” Starr wrote. “Instead, the inclusion of AI as a complement to human judgement is not only acceptable but leads to better outcomes.”
The high-impact category was established under the Trump administration’s Office of Management and Budget memo on AI governance and stems from a similar process that began under the Biden administration. Broadly, that category aims to provide agencies with a framework for mitigating the risk of cases that could impact a person’s rights or safety.
Certain uses of the technology are presumed to be high impact if they serve “as a principal basis” for agency actions but officials may still determine it doesn’t actually fall into that category. Examples of presumed high-impact uses include those related to the application for federal services, biometric identification, physical movements of robots, and — importantly for the OPM memo — federal employment determinations.
While OPM acknowledged that employment determinations are among those that are presumed high impact, it concludes that in many cases, agencies may find that it doesn’t need to fall into that category. OPM’s determinations hinge, in part, on its view that when an “agency official independently reviews supporting data, evaluates the decisional logic, and adopts the AI-assisted work product, the output generally does not serve as a principal basis for the resulting decision or action.”
That isn’t purely because human oversight exempts the use from the definition, OPM qualified, but rather that the “independent review, evaluation, and adoption supply the decision’s own justification.”
Quinn Anex-Ries, a senior policy analyst for civic technology at the Center for Democracy & Technology, said that the guidance is generally a positive step as it gives agencies clear instructions for how to go through the governance processes. Yet some areas are still worrisome.
“With any document like this, the devil is really in the details,” Anex-Ries said. “And even with the positive gains that we see with this kind of additional guidance, there are definitely some concerning things that jump out to me.”
Anex-Ries noted that agency’s “principal basis” assessment echoes a logic shared by a number of agencies in the most recent AI use case inventories. That is: Agencies have often concluded that if an AI tool isn’t directly making a decision, it isn’t considered high impact.
“What is concerning about that is that a tool can have a really significant impact on an outcome, and it may still not be considered high impact by an agency,” Anex-Ries said. The outcome is that risk management practices aren’t applied.
The “most striking example” outlined in the memo, he said, is OPM’s determination that AI used to support applicant-level evaluations generally wouldn’t be high impact.
Put more plainly, that means AI can be used broadly to assess a candidate, issue summaries, and provide recommendations about that individual that “pretty logically will have a pretty big bearing on an employment decision — and still that tool is not subject to these minimum risk management practices,” Anex-Ries said.
He emphasized that high impact doesn’t mean that a tool shouldn’t be used. Rather, it means that when that tool is used, commonsense safeguards should be applied.
And interestingly, OPM’s memo includes the same sentiment: “Agencies should aim to streamline the processes where possible. Agencies are reminded that a high-impact designation is not a prohibition on adoption.”
OPM told agencies that it will continue introducing features across tools like USAJOBS, USA Staffing and USA Hire, but said not to wait for those tools before adopting AI consistent with the memo.