"Use of Artificial Intelligence in the Federal Hiring Process" is a fourteen-page memorandum dated August 27, 2026 from Scott Kupor, Director of the Office of Personnel Management, to heads of departments and agencies, copied to Chief Human Capital Officers, Deputy CHCOs, Human Resources Directors, Chief Information Officers, Chief Artificial Intelligence Officers and Chiefs of Staff (Source: opm.gov).
Its opening proposition inverts the usual framing of AI risk guidance: "Agencies may be compromising efficiency and quality in Federal hiring by failing to adopt the use of Artificial Intelligence (AI)." The memorandum is positioned as an interpretive aid rather than a new rule. Footnote 2 states that nothing in it "is intended to impose requirements that are more restrictive than, or that otherwise modify, the requirements of OMB M-25-21," that it is "intended to clarify the interpretation and application of M-25-21 in the Federal hiring context and to support agency adoption of AI," and that in the event of any inconsistency "M-25-21 controls." Footnote 3 states that it applies to internal Executive Branch management and creates no right or benefit enforceable at law or in equity.
Authorities invoked
Section I cites Executive Order 14170, Reforming the Federal Hiring Process and Restoring Merit to Government Service (January 20, 2025), and the Merit Hiring Plan (White House Domestic Policy Council and OPM, May 29, 2025), as requiring agencies to improve federal hiring by developing better position descriptions and assessments, increasing efficiency, and reducing average time-to-hire to 80 days. Footnote 1 also cites Executive Order 14179, Removing Barriers to American Leadership in Artificial Intelligence (January 23, 2025), and the Chance to Compete Act of 2024, Pub. L. No. 118-188. AI is described as "an important tool to help implement these goals," to be leveraged in concert with OMB M-25-21, Accelerating Federal Use of AI through Innovation, Governance, and Public Trust (April 3, 2025).
The high-impact standard and OPM's reading of "principal basis"
Section II restates M-25-21's definition: a use of AI is high-impact if the AI's "output serves as a principal basis for decisions or actions with legal, material, binding, or significant effect on: (1) an individual or entity's civil rights, civil liberties, or privacy; (2) an individual or entity's access to education, housing, insurance, credit, employment, and other programs; (3) an individual or entity's access to critical government resources or services; (4) human health and safety; (5) critical infrastructure or public safety; or (6) strategic assets or resources, including high-value property and information marked as sensitive or classified by the Federal Government." Access to employment is the limb that engages in the hiring context.
OPM's interpretive move is on the "principal basis" element: "where an appropriate 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." The memorandum is explicit that the reason is not the presence of a human: "That is not because human oversight, by itself, exempts a use from the high-impact definition. Instead, it is because independent review, evaluation, and adoption supply the decision's own justification." It adds, "Agencies should not treat this analysis as a barrier to adoption."
Determination of the terms or conditions of federal employment — "including pre-employment screening, reasonable accommodation, pay or promotion, performance management, hiring or termination, or recommending disciplinary action; reassignment of workers to new tasks or teams" — sits in M-25-21 § 6's presumed-high-impact list. Footnote 5 notes that "not all AI use cases for hiring fall under this presumption." Where a category is presumed high-impact, agencies must treat it as such unless they determine the use does not actually fit the definition, in which case "an appropriate agency official must submit written documentation to notify the CAIO [Chief Artificial Intelligence Officer]." For uses that remain high-impact, agencies must either implement minimum risk management practices or submit a waiver request to the agency CAIO for one or more of them.
Footnote 4 records the textual hook OPM relies on, quoting M-25-21's footnote 27: AI "may be integrated in decision or activity pipelines in high-impact categories without meeting the definition of high-impact because the AI's output does not actually 'serve as a principal basis for' the relevant type of agency action or decision."
Four categories treated as generally not high-impact
Section III names four uses that, after the Appendix A assessment process, OPM considers generally not high-impact:
- Content development for job opportunity announcements — drafting position descriptions, evaluation statements, announcement text, job analyses, qualification requirements, crediting plans and interview questions, "when the process of drafting content itself does not affect an individual's access to employment." Before operational use the material must be reviewed, validated and adopted under classification requirements at 5 U.S.C. Ch. 51, job analysis and validation under 5 CFR 300.103, and subject-matter-expert involvement — obligations that "apply independently of M-25-21."
- Decision-support for applicant-level evaluations — résumé screening, minimum-qualifications and eligibility review, and candidate summarization "when the AI output is not a principal basis for a decision or action," the worked example being AI that locates, extracts and summarizes documentation with a reference back to the underlying record while the deciding official independently reviews that record and decides on it.
- Pre-offer quality control — review of the staffing case file and selection before a final job offer, "when the AI is limited to administrative completeness and quality-control support and is not a principal basis for approving, delaying, or rejecting the selection or final offer."
- Aggregate hiring program evaluation — analysis of completed hiring actions, applicant flow, time-to-hire and assessment performance, "where no individual employment decision turns on the AI output."
Agencies "must still assess their specific workflow and deployment context to affirmatively reach a non-high-impact determination," and an appropriate official must document it to the CAIO.
What a non-high-impact determination does not change
The memorandum states that "a non-high-impact determination changes nothing else." AI-assisted steps remain subject to job analysis and validation under 5 CFR 300.103, and agencies must verify before deployment and periodically that:
- AI-assisted rating, ranking and referral correctly apply veterans' preference;
- AI-enabled steps meet Rehabilitation Act accessibility and reasonable-accommodation requirements;
- applicant data is processed only in agency-authorized tools with applicable Privacy Act and privacy-impact-assessment protections;
- each hiring action remains documented and auditable; and
- a reconsideration avenue remains available for adverse eligibility determinations.
Footnote 6 clarifies that the fifth item "creates no new process or right," restating avenues that exist independently — correction procedures under agency delegated examining authority and veterans'-preference redress under 5 U.S.C. § 3330a — and requiring "only that AI-assisted steps not impede access to them."
The memorandum adds that "a high-impact designation is not a prohibition on adoption," and encourages Chief Human Capital Officers to collaborate with CAIOs, Chief Privacy Officers, Chief Information Officers and Chief Data Officers on the minimum risk management practices of M-25-21 § 4(b).
Appendix A: the three-step framework
Appendix A, Table 1 works seven use cases through three columns — whether the use is presumed high-impact or meets the definition; whether the AI is being used as a principal basis; and other relevant considerations. Its framework row states the interpretive rule in its sharpest form:
"The inquiry is functional, not formal: identify what in fact supplies the justification for the decision or action in the documented workflow. An AI output may serve as a principal basis even where a human formally approves the result; conversely, an output that merely informs a decision that is independently justified by the record does not."
and
"A high-impact determination is possible whether there is or is not human oversight for the decision or action. Asserted human validation or review does not, by itself, remove a use from the definition; what matters is whether the documented record shows that the official's independent judgment, rather than the AI output, supplied the decision's justification."
| Use case | Step 1 | The condition that keeps it out of high-impact — and what tips it back in |
|---|---|---|
| Draft position descriptions, evaluation statements, job announcements | May fall under the presumption; OPM would typically not consider it high-impact | Not a principal basis where the draft is a starting point the responsible official independently reviews, revises and adopts, aligned with the classifier's authority and the OPM Qualification Standards Handbook. Where an official adopts AI-drafted material "without substantive, documented review, the draft supplies the justification for what is adopted, and the use should be treated as high-impact." |
| Draft job analyses, qualification requirements, crediting plans, interview questions | May fall under the presumption; typically not high-impact | Because these instruments later govern how applicants are evaluated, independent validation under 5 CFR 300.103 against the position's actual duties and documented job-relatedness "is the safeguard that keeps the AI draft from serving as a principal basis for downstream applicant determinations." Where that validation is not performed and documented before use, treat as high-impact. |
| Screen résumés against minimum qualifications; review documents for veterans' preference, ICTAP, Schedule A, military-spouse, selective-service and citizenship eligibility; recommend eligibility determinations | Presumed high-impact (pre-employment screening, hiring, reasonable accommodation) | Not a principal basis where the AI locates, extracts and summarizes the record against the eligibility criteria; an authorized official reviews the documentation including but not limited to the AI summary and the entire package and decides on that basis; and the official justifies the determination by reference to the underlying record, such that they would reach the same result on the source documents. "Where officials in practice ratify AI outputs without independent review of the underlying record, the output serves as a principal basis notwithstanding the nominal review step." Conditions: outputs identify the specific documentation each determination rests on; documentation remains accessible to the deciding official; accuracy is established through sampling and quality assurance, "with particular attention to determinations that would remove an applicant from consideration." |
| Generate proposed recommendations of applicants who do not meet minimum qualifications; assess candidates and generate match rationales for referral and selection | Presumed high-impact (pre-employment screening, hiring, pay or promotion) | Not a principal basis where the record and the official's judgment supply the justification and the rationale is traced to identifiable evidence in the application material. "Adverse recommendations warrant particular care: where the official's recorded rationale merely restates the AI-generated rationale without independent verification against the application material, the AI output supplies the justification, serves as a principal basis, and the use should be treated as high-impact." In practice the rationale must cite the specific application material and job-analysis factors, and the selecting official's independent conclusion must be recorded in the file. |
| Rate, rank or score candidates against the adopted crediting plan; AI-scored assessments producing scores or rank order, with or without human review and certification, including AI as sole scoring instrument | Presumed high-impact (pre-employment screening, hiring) | Not a principal basis where, among other conditions, "each score traces to identified plan elements and the applicant's own responses such that an authorized official can review the derivation — including in summary form — confirm that the score is the one the adopted plan yields, and certify the result on that basis." Where scores cannot be traced to the adopted plan and the applicant's responses, where the derivation is not reviewable by the certifying official, or "where AI operates as the sole scoring instrument without a reviewable basis for certification, the score is itself the underpinning of the decision, no independent record supports it, and the output likely serves as a principal basis whether or not a human official certifies the result." |
| Review staffing case file and selection before a final job offer to identify missing documentation, inconsistencies, procedural errors or compliance concerns | May fall under the presumption; typically not high-impact | Not a principal basis where the output informs an HR professional that a package may have missing information, any flag that could remove a candidate is verified against source documentation, and the output does not influence ranking, rating or eligibility. Two conditions: "the absence of a flag may not substitute for a required certification by the authorized official," and any flag that would remove a candidate must be verified against source documentation before action. |
| Evaluate completed hiring actions and recruitment outcomes; analyze applicant flow, time-to-hire, certificate and selection patterns, assessment performance; produce quality-of-hire analytics; assess consistency with merit system principles; identify adverse impact or anomalies in aggregate data | May fall under the presumption; typically not high-impact | Outputs are "statistical and programmatic: they describe completed hiring actions and aggregate patterns rather than any pending application." If aggregate outputs are later repurposed to inform a decision concerning an identifiable individual, "that is a distinct use of AI that requires its own assessment under this framework." Aggregate analyses should use the minimum applicant data necessary; an apparent error in an individual completed action "should be referred through existing correction, audit, or oversight processes rather than acted on directly from the AI output." |
The table's assessments "reflect OPM's views on the application of OMB M-25-21 to common Federal hiring use cases, developed in consultation with OMB," and are "not official OMB determinations"; agencies remain responsible for their own determinations.
Appendix B and next steps
Appendix B supplies model determination language for the written documentation to the CAIO required by M-25-21 § 4(a), with branches for a presumed-high-impact use that is not a principal basis, one that is, and an example waiver justification. It directs that a determination that a presumed high-impact use does not meet the definition (M-25-21 App. 14) and a request to waive minimum risk management practices (App. 15) "serve different purposes and should be prepared and submitted as separate documents, not combined." The model waiver text asks the drafter to identify the specific practice, state that applying it would "increase risks to safety or rights overall / create an unacceptable impediment to critical agency operations," list compensating controls such as human review and certification, sampling and audit logs, and performance monitoring, and set a scope and duration reassessed "upon material change to the tool, workflow, or deployment context, and at least annually." Determinations are to be revisited on material change "or if testing or audit results show that the review controls described above are not functioning as documented."
Section IV states that OPM will continue introducing AI features across the USA Suite — USA Staffing, USAJOBS, USA Hire, USA Class and USA Performance — but that "agencies should not wait for those features to begin adopting AI consistent with this memorandum." OPM will also address AI-assisted processes in delegated examining oversight and assist agencies on request through the Artificial Intelligence Governance and Adoption Program (AI GAP).
Provenance
Fetched directly from opm.gov/chcoc, the Chief Human Capital Officers Council path on the issuing agency's own domain, on August 29, 2026 (HTTP 200, 14 pages). The document header confirms the addressee, signer, date and subject. Two independent accounts — Nextgov/FCW (August 28, 2026) and FedScoop — name the same memorandum, date and high-impact conclusion, and the distinctive phrase they quote appears at § II.
A capture note: an earlier markdown conversion returned Appendix A's Table 1 with cell content scrambled across columns and rows, and the ingest was deferred rather than saved in that state. The raw file now records Table 1 from a cell-boundary-preserving table extraction, rendered one row per subsection.
An attribution to keep distinct: the figure of an 87-day average federal time-to-hire, carried in coverage of this memorandum, comes from a separate OPM blog post by Adam Starr, OPM's chief information officer, not from this document. The memorandum itself gives no such figure; it cites the Merit Hiring Plan target of reducing average time-to-hire to 80 days.
Relationships
- depends-on: OMB M-25-21 — Accelerating Federal Use of AI through Innovation, Governance, and Public Trust — the high-impact standard this memorandum interprets and expressly defers to.
- related: OPM — Office of Personnel Management (AI Deployer) — the issuing agency.
- supports: Federal AI Compliance Landscape — an agency-level interpretation of a government-wide requirement.
- related: AI Bias and Discrimination — the substantive risk the high-impact designation is directed at in employment.
- supports: Meaningful Human Review — supplies a functional rather than formal test for when human review displaces the AI output as the decision's justification.
- related: Human Oversight.
- related: Government AI Procurement — adoption guidance for federal AI tooling.