Federal Hiring AI Rules: What Scans You, What You Can Use

Federal hiring is moving in two directions at once. On 27 August 2026 the Office of Personnel Management told agencies to widen their use of AI across the hiring process, and reporting a week later described AI agents conducting screening interviews for the Tech Force program. Meanwhile the government's own application instructions ask candidates to certify they did not use a large language model to write their answers, and at least one department discontinues candidacies over it.

That asymmetry is the practical situation for anyone applying to a federal technical role right now. This article sets out what the government has authorised on its side, what it forbids on yours, and which of those rules carry real consequences. Facts are current as of 8 September 2026. It is a reading of public memoranda, agency instructions, and news reporting, not legal advice.

What OPM authorised on 27 August 2026

OPM Director Scott Kupor issued a memorandum to heads of departments and agencies titled Use of Artificial Intelligence in the Federal Hiring Process (OPM CHCOC memo, 27 August 2026). It identifies four uses OPM considers generally not high-impact:

• drafting job announcements, position descriptions, and evaluation statements;
• decision support for applicant evaluation, including résumé screening and qualifications review;
• pre-offer quality control on administrative completeness;
• aggregate analysis of completed hiring actions.

The condition attached to all of them is human review that is real rather than nominal. The memo's own framing is that "the official's independent review and adoption — rather than the AI draft — supply the justification" for a decision. It adds the converse: where officials "in practice ratify AI outputs without independent review of the underlying record," the AI output becomes a principal basis for the decision, which pulls the use back into the high-impact category and its heavier obligations.

The memo also imposes checks that apply regardless of classification. Agencies must verify that AI-assisted steps correctly apply veterans' preference, meet accessibility and reasonable accommodation requirements, process applicant data only through authorised tools under Privacy Act protections, remain "documented and auditable," and preserve "a reconsideration avenue for adverse eligibility determinations."

What the memo does not do is require anyone to tell you that AI was involved. Reporting on the memo noted no applicant-notice provision, and the guidance is addressed to agency officials rather than to candidates (FedScoop, 28 August 2026). OPM separately said AI features are coming to USAJOBS, USA Staffing, and USA Hire, and encouraged agencies not to wait for them (Federal News Network, August 2026).

The Tech Force pilot and what it captures

CBS News reported on 3 September 2026 that AI interviews were being rolled out for some federal hires, beginning with Tech Force — the OPM program that recruits private-sector technologists into two-year government posts. The described setup is early-stage structured interviews with standardised questions, conducted by AI agents, with audio recorded and summaries generated for hiring managers to review afterwards (CBS News, 3 September 2026). Subsequent coverage identified the vendor as CodeSignal and described phone, audio, and text formats, with transcribed recordings forwarded to hiring managers (Jerusalem Post, 7 September 2026).

Both accounts rest on unnamed sources rather than a published OPM announcement, so treat the operational detail as reported rather than confirmed. What is on the record is the vendor relationship: CodeSignal's own customer page states it has worked with OPM since 2025 on technical skill assessments for Tech Force, covering software engineers, data engineers, product managers, and cybersecurity roles, and claims roughly 80% of candidates who passed its evaluations went on to pass screening interviews (CodeSignal OPM customer story). That figure is a vendor claim about its own product and has not been independently verified.

If you are sent one of these, the mechanics are the same ones that apply to CodeSignal's commercial product — the question set, the transcript, and what reaches a human reviewer. Our breakdown of what CodeSignal's AI Interviewer records and scores covers the capture behaviour in more detail than any federal document currently does.

Why the "high-impact" label matters to you

The classification is not bureaucratic housekeeping. OMB Memorandum M-25-21, issued 3 April 2025, defines high-impact AI as AI whose output "serves as a principal basis for decisions or actions with legal, material, binding, or significant effect," and it explicitly lists Federal Employment Terms — "including preemployment screening, reasonable accommodation, pay or promotion, performance management, hiring or termination" — as a presumed high-impact use case (OMB M-25-21).

Systems in that category owe you specific things. M-25-21 requires agencies to "offer consistent remedies or appeals," under which "affected individuals should have access to a timely human review and remediation process," and to provide additional human oversight with appropriate fail-safes. An agency that cannot meet the minimum practices must seek a waiver; OPM's own compliance plan describes a board that evaluates waiver requests case by case and may approve, deny, or conditionally approve them (OPM AI Compliance Plan, September 2025).

The August 2026 memo does not repeal any of that, but it does describe a route by which AI-assisted screening stays outside the high-impact category: if a human independently reviews the underlying record, the AI is decision support rather than a principal basis. Policy analyst Quinn Anex-Ries criticised exactly that logic in FedScoop's coverage, noting that a tool "can have a really significant impact on an outcome" while still escaping the classification, and flagging applicant-level evaluation tools as the concerning case. Whether the safeguards reach a given candidate therefore depends on how a specific agency documented a specific use — which is why the question to ask is about the use case, not about AI in general.

The rules pointing the other way

While agencies expand AI use, applicants are being asked to disclaim it. OPM's instructions for the merit hiring essay questions tell candidates they "will be asked to certify that you are using your own words and did not use a consultant or artificial intelligence (AI) such as a large language model (LLM) like ChatGPT or Copilot" (FedWeek summary of OPM guidance). Responses are capped at 200 words each and, per the same guidance, are not scored or rated.

Some agencies go further. The State Department's Foreign Service selection FAQ states that "candidates are prohibited from using any artificial intelligence (AI) tool, to include but not limited to ChatGPT, to aid in their written responses," and that the Department "will discontinue any individual's candidacy if found to have violated this prohibition" (State Department FSO selection FAQ).

The consequence structure is also different from private-sector hiring. Federal applicants sign the Declaration for Federal Employment, Optional Form 306, which warns that "a false or fraudulent answer to any question or item on any part of this declaration or its attachments may be grounds for not hiring me, or for firing me after I begin work, and may be punishable by fine or imprisonment," citing 18 U.S.C. § 1001 (OF-306, revised October 2019). A certification you signed falsely is a different category of problem from a platform terms-of-service violation. That distinction is worth holding onto before treating federal assessments like commercial ones — the general private-sector picture in whether you can use AI during a technical interview does not transfer here.

What you can reasonably ask for

Federal applicants have some concrete handholds that private-sector candidates lack.

Ask which assessment you are being sent, by name. The reconsideration avenue the OPM memo requires attaches to adverse eligibility determinations. You cannot invoke it without knowing what produced the determination.

Check the public inventory. Agencies publish AI use case inventories, consolidated annually by OMB (2025 Federal Agency AI Use Case Inventory). If an agency has registered a hiring-related use case, it is on the record with a stated risk classification.

Use the Privacy Act. The memo requires applicant data to be processed only through authorised tools under Privacy Act protections and records to remain documented and auditable. That is the basis for requesting your own record.

Raise accommodation needs early. Accessibility and reasonable accommodation compliance is one of the requirements OPM applies to every AI-assisted step, high-impact or not.

Asking for a human interviewer is a weaker request than these, in federal hiring as elsewhere — the realistic options when an employer sends an AI-led interview apply, and the answer is usually goodwill rather than entitlement. Note also that the transparency rules with real teeth in 2026 are European and US state-level; nothing in the federal memoranda above obliges an agency to tell a candidate that AI screened them.

Where this leaves you

If you are applying to a federal technical role this autumn, assume an AI system touches your résumé and may conduct your first interview, assume nobody will tell you, and assume the safeguards exist on paper at the agency level rather than as rights you can assert on the spot. The most useful thing you can do is name the assessment and ask what happens to the recording.

On the other side of the ledger, the federal application asks for a certification that private employers do not, backed by a statute that private employers cannot invoke. Control is a desktop AI interview assistant, and we would rather say plainly that federal assessments carrying a signed declaration are not the place to test one. Use the preparation time instead: our guide to running an AI mock interview with a scoring rubric is a better fit for structured, standardised question sets of the kind Tech Force screening reportedly uses.

Reporting on the Tech Force pilot is a week old and rests on unnamed sources. If OPM publishes an announcement or an updated use case inventory entry, the picture here will get firmer, and some of it may change.