Proctoring and flagged activity with AI Interviewer
Overview
Proctoring in AI Interviewer is designed to detect and log suspicious behavior during automated interviews, so recruiters can review potential misuse of unfair means after the fact. It helps protect interview integrity, especially in remote or unmonitored settings, without automatically rejecting candidates or changing their scores.
Proctoring in AI Interviewer gives customers visibility into what happened during an interview, not just what was said. By:
- configuring the right mix of checks in Proctoring checks, in Admin Console ▸ Product Setup ▸ AI Agents,
- training reviewers to interpret flags as prompts for review, not automatic failures, and
- aligning proctoring choices with internal policy and candidate context,
organizations can use this feature to strengthen fraud detection and build trust in AI-led interviews, while keeping humans firmly in control of hiring decisions.
Two layers decide what runs. Your admin turns on the checks your organization allows, and a modular AI interview guide can turn any of them off for its own interview. What a guide can't do is add a check the admin left off.
To know more about setting proctoring checks per guide, read Guide-level proctoring.
What proctoring captures
During an interview, the system monitors for specific behaviors and logs them as flagged activities. These signals appear to the recruiter under Flagged activities in the interview feedback view, alongside the recording and transcript.
Key characteristics:
- Each enabled check becomes a signal that can be triggered for an interview.
- Recruiters see timestamped flags and can go directly to the relevant moment in the recording and transcript.
- Recruiters can manually flag or dismiss events if they believe the signal is a false positive or contextually acceptable.
- Depending on the check, candidates may see a real-time alert when something is detected. Flags reach reviewers either way.
- Proctoring doesn't change the AI score. Most checks only record what happened, but the Restrictions checks can also alert the candidate in real time or block the behavior outright—see the tables below.
Current set of proctoring checks
Below is the current set of checks available for an AI Interviewer, grouped the way they appear across the product. The names shown here are the ones a reviewer sees in the interview feedback view.
Camera and audio
| Check | What it detects | Why it matters | Notes for reviewers |
| Microphone muted | Candidate turns their microphone off and on; frequent toggles are logged. | May indicate side conversations or coaching; also surfaces audio issues. | A few toggles are normal; review if muting is frequent or during critical questions. |
| Moved out of view | No part of the candidate's head is visible in the webcam frame when video is enabled. | May indicate the candidate stepped away to consult resources or seek off-camera assistance. | Short absences can be benign; repeated or long absences during answers warrant review. A candidate who turns away or looks down still counts as present, so this flag now means they genuinely left the frame. |
| Unknown speaker was detected | The voice on the recording doesn't match the candidate. | Suggests someone else is answering, either in the room or remotely. | Check the moment against the recording—a brief cross-talk from a shared space reads differently from a sustained second voice. |
| Multiple people in view | More than one person appears in the camera frame when video is enabled. A person visible in a mirror counts as a second person. | Designed to detect external assistance or proxy candidates. | Brief passersby in shared spaces can create false positives; a second person consistently in view is a stronger concern. A reflection of the candidate is flagged deliberately, so the same reflection is read the same way every time. Dismiss it on review. |
| Prohibited item seen on screen | A phone, laptop, or book in the candidate's hands or within arm's reach. An item further away, such as a bookshelf across the room, isn't flagged, and neither is an object reflected in a mirror. | Physical reference material and secondary devices are common cheating vectors. | Consider the role and environment; focus on when the item appears and whether the candidate visibly interacts with it. A phone held to the ear, or held low with only its top edge in frame, is now identified. |
| Face mismatch detected | The face on camera changes during the interview. | A different person taking over mid-interview is the clearest form of impersonation. | Correlate with the recording; lighting shifts and a candidate moving in and out of frame can trigger it. |
Session activity
| Check | What it detects | Why it matters | Notes for reviewers |
| Browser tab switched | Candidate switches away from and back to the interview tab, based on browser visibility and focus. | One of the clearest indicators that a candidate may be looking up answers or referencing external material. | Single brief switches might be network- or UX-related; multiple switches during problem-solving deserve closer review. |
| Rejoined interview | Candidate drops and rejoins the interview repeatedly, including frequency and timing. | Frequent rejoins can indicate pauses for coaching or external research. | Correlate with other flags, for example tab switches or developer tools. |
| Developer tools accessed | Candidate opens browser developer tools during the interview. | Developer tools can be used to inspect page content or network calls, which is atypical for standard interviews. | In non-technical roles, any use is unusual; for highly technical roles, align interpretation with your internal policy. |
Restrictions
These checks can block the behavior, not just flag it. Whether blocking is on is your admin's decision.
| Check | What it detects | Why it matters | Notes for reviewers |
| Fullscreen exited | Candidate enters or exits fullscreen mode on the interview screen. | Leaving fullscreen can precede switching windows or bringing other content into view. | Occasional exits can be benign; repeated exits plus other flags are more suspicious. When blocking is on, the interview pauses and the mic mutes until the candidate returns to fullscreen. |
| External monitor detected | A second display is connected during the interview. | A second screen can hold reference material out of camera view. | Blocking pauses the interview until the extra display is disconnected. Requires a supported desktop browser. |
| Screen sharing stopped | Candidate stops sharing their screen during the interview. | Ending the share removes the reviewer's view of what else is on screen. | Blocking pauses the interview until the candidate shares again. Requires a supported desktop browser. |
| Copy or paste attempted | Copy, cut, and paste attempts during the interview. | Pasted answers and copied prompts are a common shortcut, particularly on written or coding questions. | Blocking prevents the action outright, so a flag with blocking on means the attempt failed. |
| Keyboard shortcut attempted | Use of a shortcut on the blocked list, for example a page reload or an inspect shortcut. | Some shortcuts open developer tools, reload the page, or reach the address bar. | Your admin sets which shortcuts are on the list. Blocking prevents them during the interview. |
Note: A few activities aren't configurable checks and always appear on the feedback page, such as focus mode being enabled and moving to the next section. They aren't listed above because there's nothing to turn on or off.
Candidate experience and privacy considerations
From a candidate's perspective, proctoring is part of a broader trust and transparency experience:
- Before joining the interview, candidates see clear disclosures and consent screens explaining that the interview is recorded and monitored for integrity.
- Some checks depend on video being available, for example moved out of view, multiple people, and prohibited items.
- For some checks, candidates see a real-time alert when something is detected, and the Restrictions checks can pause the interview or block the action until the candidate resolves it. Which checks behave this way is your admin's decision.
- For customers who prefer audio-only review for fairness reasons, the system can hide the video player for recruiters while still running proctoring checks in the background and surfacing flags.
- Proctoring data isn't used for video analysis beyond the documented checks; final decisions remain with human reviewers.