LunaPromptsThe assessment score is strong. A session alert says something deserves attention. The recruiter now has two pieces of information that seem to disagree, and an interview panel waiting for a recommendation.
This is where an integrity process earns its place. The next step should produce a clearer account of the work, including the candidate’s explanation. A label without that account gives the team very little to act on.
LunaPrompts helps hiring teams investigate possible unauthorized AI assistance during assessments through session activity, authorship indicators, and follow-up explanations. Integrity evidence sits alongside performance so reviewers can understand the circumstances behind a result.
Review a pasted implementation
Technical reviewerIllustrative scenario01 / 03A large paste appears
A working function arrives in one edit. The session timeline marks the event.
A candidate submits a retry helper in one large paste. The task allows a coding assistant but requires the candidate to review the output. The interviewer points to the retry condition and asks what happens when the first request succeeds but its response is lost.
A useful answer identifies the risk of a duplicate operation and connects a proposed change to that failure. Ask the candidate to adjust the example or write a check. The evidence is the explanation and modification, rather than the confidence of the delivery.
Now change the original rule: imagine the same task explicitly prohibited outside assistance. The same technical explanation would demonstrate understanding, but it would not by itself resolve whether the rule was followed. Record those as separate questions. This prevents technical competence from erasing a policy issue, or a session anomaly from erasing demonstrated competence.
A practical review note has four parts: the rule candidates received, the observed event, the candidate’s account, and the reason for the reviewer’s conclusion. If an important fact remains unknown, keep it unknown in the report instead of filling the gap with an accusation.
Start with a clear rule about AI use
Some roles should be assessed with AI tools available. Other tasks are intended to establish independent coding or conceptual ability. Both can be useful, but the rules need to be explicit.
Tell candidates which tools are allowed, what work must be their own, and what session information will be collected. A permitted coding assistant is part of an AI-assisted assessment. Undisclosed assistance in a task that prohibits it raises a different question.
Review several kinds of integrity evidence
LunaPrompts brings together signals that can help reviewers decide what deserves a closer look:
- Session activity: Examine relevant events such as tab or window changes, paste activity, and typing patterns in their time context.
- Authorship indicators: Review similarity and other available evidence about the relationship between a submission and the candidate's work.
- Identity checks: Use the identity and face-verification capabilities available in the selected assessment setup.
- Candidate explanations: Ask the candidate to explain a decision or adapt part of their submission through AI Viva or a human follow-up.
- Enhanced environment signals: Selected Secure assessments can add native-environment checks beyond the browser session.
The platform's anti-cheating models help surface activity for investigation. A flag is an input to the review, not a finding that a person cheated.
What about Cluely, ParakeetAI, and similar tools?
Hiring teams ask about real-time assistance tools by name because they want to know whether an apparently independent answer could have external help. LunaPrompts' integrity workflow addresses indicators of that assistance. Detection coverage depends on the environment, enabled checks, and the way the tool is used.
It would be misleading to promise that every tool, version, or technique can always be detected. The useful hiring outcome is a documented review: what happened, why it was flagged, and how the candidate explained their work.
Make integrity review part of a consistent process
Agree on the assessment's tool policy before invitations go out. Review the supporting evidence for a flag. Consider technical interruptions, accessibility needs, and the candidate's explanation. Record why the team reached its conclusion.
That process gives recruiters and engineering managers a shared basis for discussion. It also helps keep an integrity concern from becoming an unsupported label attached to a candidate.
A review framework your team can use
The following is an editorial checklist for this workflow. Adapt it to the role, the candidate journey, and your configured environment.
| Review area | What to establish |
|---|---|
| Tool policy | The exact instructions sent before the task |
| Observed event | A timestamp and the relevant work |
| Follow-up | A specific question or small adaptation |
| Conclusion | What is supported, unresolved, or explained |
Do not turn a detector score into a claim about a named tool. If the environment cannot establish which application was used, the report should not name one. Ask the vendor what the signal measures and what supporting evidence a reviewer can inspect.
Put the guide to work
Take one example flag and write the candidate-facing follow-up before looking at the overall score. Check whether a second reviewer can distinguish observations from interpretations in your note.
Download the editable worksheet to record the decisions and open questions with your team. The examples in this guide are illustrative, not customer results or validated selection rules. Product availability and setup depend on the workflow agreed for your account.
A focused checklist, a worked scenario, and questions to resolve together.
Frequently asked questions
Does AI-generated code automatically mean cheating?
No. AI-generated code may be permitted in the assessment. The relevant questions are whether the candidate followed the stated rules and demonstrated the required ability.
Can an integrity score prove that a candidate used a specific tool?
A score or indicator alone should not be treated as proof of a named tool or misconduct. Review the underlying evidence and the capabilities of the assessment environment.
Does this replace a technical follow-up?
No. A focused follow-up can clarify both understanding and authorship. Integrity signals help the team decide where that conversation would be useful.
Keep building your hiring workflow
- AI Viva: They have the answer. Ask what would change it.
- Candidate analytics: The scores match. The follow-ups should not.
- Interview copilot: They already took the test. Make the interview add something.
Explore assessment integrity with LunaPrompts or browse all 19 hiring field guides.
