Understanding interview integrity violations
The AI flags moments during the interview that suggest possible integrity issues — third voices, frequent screen switches, suspicious pauses. Here's how to read them without overreacting.
Key takeaways
- The Violations panel on the Interview Review page lists timestamps where the AI flagged something unusual.
- Common flags: third voice detected, candidate looked away repeatedly, browser tab/window switch, prolonged silence.
- Each flag is a *signal*, not a verdict. Most are benign — verify in the recording before concluding misconduct.
- Patterns of flags across multiple moments are stronger signal than a single isolated flag.
Before you start
Every completed interview goes through an integrity check that flags moments the AI considers unusual. These flags show on the Interview Review page in a Violations panel. This article covers how to read them without making bad inferences.
What gets flagged
- Third voice detected — the AI heard speech that wasn't the candidate. Could be a family member in the next room, a podcast playing nearby, or, rarely, someone feeding the candidate answers.
- Browser/tab/window switch — the candidate's focus shifted away from the interview tab. Could be a notification, an alt-tab habit, or a parallel Google search.
- Prolonged silence followed by fluent answer — the candidate paused for an unusually long time, then delivered a polished answer. Could be thinking deeply, or could be reading from notes.
- Repeated eye-darting — the candidate's eyes moved away from the camera in a pattern consistent with reading from another screen.
- Audio anomaly — sudden volume change, microphone change, or compression artifact suggesting the audio source switched.
How to verify a flag
Patterns vs isolated incidents
TIP
A single 'third voice' flag at minute 12 of a 45-minute interview is almost certainly background noise. Five 'third voice' flags clustered around the hardest technical questions is a different signal entirely.
When making a judgement, look for patterns: do flags cluster around hard questions? Do they consistently align with the strongest answers? Or are they spread evenly throughout, including during small-talk moments? Even distribution = noise. Clustering at high-stakes moments = worth taking seriously.
What to do when violations are real
- Re-issue the interview with a clearer pre-flight reminder about being alone in the room.
- Reject the candidate for this role. The AI evaluation may or may not be reliable in the presence of confirmed integrity issues.
- Flag to your team so others reviewing the same candidate for other roles see the context.
What NOT to do: name and shame, share the recording outside your team, or use the AI flag as a sole basis for a rejection without verifying in the recording first.
Frequently asked questions
Does a violation flag mean the candidate cheated?
No. A flag means the AI noticed something unusual. The actual reason might be: a flatmate walked into the room, the candidate dropped a pen and reached for it, a notification flashed, or, occasionally, genuine misconduct. Always verify in the recording before treating it as evidence.
How sensitive is the violation detector?
Calibrated to flag with moderate sensitivity — we err on the side of slightly more false positives over missing real issues. Most interviews have 0–3 violation flags; senior-role interviews typically have more because they're longer.
Can I disable violation detection for a specific job?
Not at the moment. It runs on every interview. If you're hiring for a role where this is inappropriate (e.g. an accessibility-focused role where candidates need assistive software), reach out to your account manager.