Your Interview Questions Are Public. That Does Not Prove Cheating.

The leak is an instrument problem before it is a candidate problem

Imagine a technical interview question that has been used for 18 months. Someone posts the exact prompt and a model answer publicly. A finalist gives an unusually polished answer. The tempting conclusion is that the finalist cheated. Yet the same answer could come from legitimate preparation, previous work, a coaching course or independent reasoning. Exposure is evidence about the assessment instrument; it is not proof about any person’s conduct. ^↗

If a question is publicly available, its ability to distinguish candidates who can solve unfamiliar work from candidates who memorized an answer may have changed. The remedy is a defensible review of question exposure, scoring and alternative evidence. APA test-security guidance identifies live items, scoring keys and calibration materials as sensitive assessment assets. ^{↗ ↗}

The three-part evidence ladder

Level 1 — confirmed exposure: a dated public URL or screenshot contains the exact prompt or answer key. Record the evidence in a restricted assessment log, not the candidate’s personnel file. Level 2 — potential score distortion: compare answer patterns and item performance by administration date and interview format, with enough observations to avoid overinterpreting small samples. Level 3 — individual misconduct: do not infer this from a polished response or a shared link; it requires separate, reliable evidence and an established review process. ^↗

Keep these levels separate in every meeting. The phrase “candidate likely saw it” collapses all three into one unsupported accusation. The ITC/ATP technology-based assessment guidance emphasizes fairness, validity and security as linked design responsibilities. ^{↗ ↗}

Rotate the stimulus, not the competency

A common mistake is replacing a leaked structured question with a harder, unrelated brainteaser. That fixes exposure by breaking comparability. Instead keep the same competency, difficulty target, allotted time and scoring anchors; develop two equivalent scenarios with subject-matter experts. Pilot them before operational use, and document which candidate cohorts saw which form. ^↗

Example: the old prompt asks a data engineer to diagnose a delayed nightly ETL job. A replacement can ask about a delayed streaming ingestion pipeline, provided both measure the same fault-isolation and stakeholder-communication behaviors. Do not claim the forms are equivalent until the pilot provides evidence. OPM recommends pilot testing and documenting uniform changes to scoring. ^↗

The legitimate-preparation boundary

Candidates should know the skills assessed, the interview format, expected duration, accessibility options and whether tools or notes are allowed. Hiding these details creates unequal preparation conditions without necessarily improving validity. A 2025 Human Resource Management study argues that practice opportunities can help equalize preparation for employment tests. ^{↗ ↗}

Publish a practice format and competency outline, not live scoring keys. If a question is deliberately open-book, say so and score reasoning, tradeoffs and verification rather than recall. If a question must remain confidential, limit who can export it and audit access to the bank. ^↗

A 48-hour containment workflow

Hour 0–4: confirm the leak and secure the item bank. Hour 4–12: identify impacted administrations, not individual suspects. Hour 12–24: decide whether to suspend that item prospectively, offer a consistent supplemental assessment, or preserve results with a documented limitation. Hour 24–48: approve candidate communications and a versioned replacement. Never quietly rescore only candidates you already dislike. ^↗

If candidates have already been assessed, a changed question can create inequity. Decide with the assessment owner and legal/HR reviewers how to treat cohorts consistently. Document exceptions and potential adverse impact. EEOC guidance stresses job-related and appropriate use of selection procedures. ^↗

The counterintuitive test: check for answer fragility

A memorized answer can sound perfect on the main prompt but collapse under a job-related follow-up asking for assumptions, failure modes, or an alternative constraint. That follow-up should be planned, consistent and scored against a defined rubric, not improvised to trap one candidate. ^↗

Operational synthesis: measure whether exposed items lose their ability to discriminate across otherwise comparable cohorts. A single impressive answer is weak evidence. A repeatable change in item behavior, combined with documented public exposure, is stronger evidence for redesigning the assessment. ^↗

A decision record that survives scrutiny

Record the item ID and version, first confirmed exposure date, affected candidate cohorts, job competency, scoring rule, review owner, mitigation chosen, and whether all affected candidates received comparable treatment. Keep access tightly restricted. Separate an assessment-security incident log from candidate evaluation records. ^↗

The final decision should answer: “What evidence can we still trust?” rather than “Which candidate should we blame?” ^↗

Run the operational check

Use the interactive worksheet below to test the decision logic with hypothetical or aggregated inputs. Keep identifiable candidate information out of this page.

The decision worth changing this week

Assign one owner, pilot the procedure on a small group, and record what evidence would falsify the hypothesis. Do not turn a process-audit score into an automated candidate rejection or a legal conclusion. ^↗