Recagn
Sign in

Guide · Anti-cheating

What a browser can detect in an online interview, and what it misses

Every signal an interview page can read, from tab switches to the webcam, with the browser feature behind it, what it catches and what it misses. A signal is a reason to ask; the decision stays with you.

Part of Interview cheating: how candidates cheat, what tools can detect, and what to do

What can a browser detect in an online interview?

A browser tells the interview page when it is hidden or loses focus, and what is pasted into it. It also reports key presses while the page has focus and, with the candidate's permission, camera frames and a shared screen. Nothing else from other apps or devices reaches the page.

This guide is the reference table behind the guide to interview cheating, which covers how candidates cheat and what to do when one is flagged.

The signals come in three kinds. Some are events the page hears about by itself: leaving the page, losing focus, a paste, a key press. Some are things the page has to ask the browser for, which the browser can refuse or the candidate can end: full screen, the list of screens, a shared screen, the camera. And one isn't a browser feature at all: a server comparing two faces.

What each signal catches and misses

SignalBrowser featureWhat it catchesWhat it misses
Leaving the page (tab, window, app)Page Visibility, the visibilitychange eventThat the page was hidden: a tab switch, a minimised or closed window, or on a phone a switch to another app, and when it came backWhy the candidate left, and what they looked at
Focus moving to another windowWindow blur and focus eventsThat the candidate clicked into another app or another screen while the page stayed visible, and for how longA window read without being clicked; what that window showed
Copy and pasteClipboard events (MDN, paste)A paste into the page, with its text and length; copies and cuts made from the pageRetyped text; dictation; text dragged in, which arrives as a drop event, not a paste
Developer toolsNone. No MDN, WHATWG or W3C page read for this guide defines one; a page hears key presses while it has focusA developer-tools keyboard shortcut pressed while the page has focusDeveloper tools opened from the browser menu, or before the page loaded
Full screenFullscreen API (MDN support data)That the page left full screen, on browsers that offer itHolding the candidate there: the spec tells browsers to keep a way out that always works, and a page can enter full screen only in response to the candidate's own input, such as a click, tap or key press. iPhone Safari has no element full screen
Second monitorWindow Management, screen.isExtended (MDN)On Chrome, Edge and Opera, and on Chrome, Opera and Samsung Internet for Android, that more than one screen is attachedWhat is on the other screen, and whether the candidate looks at it; on Firefox and Safari, whether there is one
A phone or another deviceNone found in MDN, WHATWG or W3COnly what the camera or the shared screen happens to showA phone below the camera or beside the laptop; help through an earpiece
Screen recordingScreen Capture, getDisplayMedia() (MDN, Chrome)The screen, window or tab the candidate chooses to share, to replay later; which of the three was chosenOther screens; phones. The candidate chooses what to share and can end the share, which the page hears about. Android browsers don't support it
WebcamgetUserMedia()Camera frames, with the candidate's permission. A server can read a still for faces, how many, and the head's pitch, roll and yawEverything outside the frame; with stills, everything between them
Face matchNot a browser feature: a server compares the faces in two imagesA face that differs between two images: the ID photo and a selfie, or the selfie and a later stillWhether the ID document is genuine. AWS calls its own comparison probabilistic and recommends a person review it

Source: MDN, WHATWG, W3C, WICG and Chrome documentation, and the AWS Rekognition API reference, as linked; browser support from MDN's compatibility data; read 17 Nov 2026

What can't a browser see?

A browser can't see a phone or laptop beside the computer, or what is on another screen. It also can't hold a candidate in full screen, and no browser feature reports developer tools being open.

A phone or another device

No web feature reports a phone, tablet or laptop near the computer. A page sees a phone only if the webcam or the shared screen happens to show it.

What a tool can do instead is read webcam stills for a second face, or for a head turned away across a run of stills, and let a person watch a recording. None of these sees a phone held below the camera or out of frame. The surest check is still the live round: ask the candidate to go one step deeper on an answer they gave.

A second monitor

Chrome, Edge and Opera, and Chrome, Opera and Samsung Internet on Android, can tell a page that more than one screen is attached, but not what is on it. Firefox and Safari, on computers and phones, can't tell it at all (MDN, isExtended, read 17 Nov 2026).

A page can ask for more detail about the screens, but that needs the window-management permission, which the browser may ask the candidate for (W3C Window Management, Editor's Draft, 10 Nov 2026). A click onto the other screen moves focus away from the page, and that the page does hear about. A screen recording shows only the screen the candidate shared.

Full screen

The Fullscreen standard tells browsers to keep a way out of full screen that always works, and a page can enter full screen only in response to the candidate's own input, such as a click, tap or key press. iPhone Safari offers a page no element full screen, and on an iPad it shows an exit button that can't be hidden (MDN support data, read 17 Nov 2026).

Keyboard Lock, which lets a page receive keys the browser would normally keep, works only in full screen. Some shortcuts, such as Ctrl-Alt-Del on Windows, stay with the operating system anyway. That is why some vendors ship an installed secure browser, which runs as an app on the computer rather than as a page.

Which browser signals raise false alarms?

Anything that takes focus or hides the page reads the same as a switch to a chatbot, such as a notification the candidate clicks. A still in poor light can show no face, and AWS calls its own face comparison probabilistic. So treat each signal as a reason to ask, and keep the decision yours.

The browser can't tell these cases apart. The visibility and focus events say that the candidate left and when, not why. A paste carries its text, not where the candidate wrote it. A head turned in a still is an angle, not a reason.

Weigh patterns over single events, and read each event against the answer that came after it.

Questions to ask any vendor about browser signals

Take these to any tool you are weighing, Recagn included:

  • Which browser events does it record, and on which browsers?
  • Does one tab switch count once, or twice?
  • If it says it sees a second monitor or a phone, which browser feature tells it, and on which browsers?
  • Does its screen recording work on phones, or ask for a computer?
  • Does it publish the rules that turn a signal into a concern?
  • Does any signal enter the score, or reach your client?

How Recagn does this

Recagn is our product. Recagn records activity (tab switches, copy and paste) on every interview set up since Activity Tracking shipped, and adds webcam proctoring, an ID check by face match and screen recording where you switch them on. Its published rules turn what it records into Integrity: No, Some or Strong concerns, never a number.

Here are the same signals, with what Recagn records for each and how its rules weigh it.

SignalWhat Recagn recordsHow Recagn weighs it
Leaving the page, focus movingEach tab switch and each click into another window, with how long. One switch counts once, and picking a file for a Document answer doesn't count. With the webcam on, a still is taken on return from a long stretch awayLong stretches away are highlighted; repeated or long time away adds up to Some concerns
Copy and pasteCopies, cuts and pastes, with the text (long digit runs masked). A paste outside an answer box is stopped, and only its length kept. Recagn tells a paste copied inside the interview from one brought inLong pastes from outside the page are highlighted; several add up to Some concerns
Developer toolsA developer-tools keyboard shortcut pressed while the page has focusA developer-tools keyboard shortcut is highlighted and can raise Some concerns
Full screenThe interview runs in the candidate's own browser window, with nothing to install; a switch away from it is recorded as time awayA switch to another app or window counts as time away
Second monitorScreen recording captures the one entire screen the candidate shares; a click onto another screen is recorded as time awayA click onto the other screen counts as time away; the recording is there for you to watch
A phone or another deviceWebcam stills and, with Continuous, camera and microphone video, for you to check for a phoneYou look at the stills and any video yourself. Integrity's camera rules read the faces in each still, how far the head is turned to the side and the Selfie match, as in the Webcam row
Screen recordingThe one entire screen only: a shared window or tab is refused and the candidate asked again. A stopped share brings up a cover until the screen is shared again, and the time off is recorded. Turning it on asks candidates to use a desktop or laptopA stopped share is highlighted; the screen off for long is a Strong concern. You watch the recording; Recagn's rules read page events and stills
WebcamSnapshots: stills at random gaps, plus one on return from a long stretch away. Continuous: stills more often, plus camera and microphone video. Recagn counts the faces, reads how far the head is turned to the side and, with the ID check on, compares each still with the SelfieStrong: a second face, or the camera off for long. Some: no face, or the head turned away, in a run of stills; the camera off briefly
Face matchID check by face match: before Start the candidate photographs an Aadhaar, PAN, passport, driving licence or voter ID and takes a Selfie with the camera. Recagn compares the two faces, then each still with the SelfieAn ID photo or a still that doesn't match the Selfie is a Strong concern; an ID that couldn't be read is a Some concern. The result never stops a candidate from starting

Source: Recagn's activity, proctoring and Integrity code, rule table integrity@1, checked 17 Nov 2026 · each rule's level and threshold: Recagn's Integrity rules

Strong concerns come from the camera, the ID check or the screen; Some concerns come from activity and lighter camera signals. Each rule, its level and the check it needs are on Recagn's Integrity rules. A highlighted event is a place to look, not a finding. A finding is a rule that fired, written as one sentence on the interview's timeline.

The finding on Farhan I.'s Interview timeline

Find the mark at 2:05: the finding opens at its moment, beside the Question that was on screen.

Farhan I. · Interview timeline

Questions
  1. Q1
  2. Q2
  3. Q3
  4. Q4
  5. Q5
  6. Q6
Activity
Stills

0:005:50

2:05 · Q2 · Text Strong

The webcam still at 2:05 didn't match the Selfie.

Given an orders table and a users table in PostgreSQL, write a query to find the top 5 users by total order value in the last 30 days, and explain which indexes would make it fast.

Recagn's rules list what happened; you decide what it means. Not used in scoring or ranking; the hiring company never sees it.

From a real Integrity run, 17 Nov 2026 · Backend developer (Node.js) Interview with Webcam Snapshots and Face match · the candidate was performed for the example

This is Farhan I.'s interview timeline, from a run on 17 Nov 2026 with a performed candidate. The role is Backend developer (Node.js), with webcam Snapshots and the ID check by face match switched on. Read it lane by lane, as the browser signals in the first table.

Activity. Each stretch away is the page being hidden: the visibilitychange event from the first row, with how long it lasted.

Stills. Each mark is a camera frame from getUserMedia(), analysed on Recagn's side after the interview, when it is scored.

The mark at 2:05. It isn't a browser event. It is Recagn comparing one still with the Selfie, and the finding reads: "The webcam still at 2:05 didn't match the Selfie." It sits in question 2, inside a stretch away that began at 1:30.

What each lane answers. The browser reported that the page was hidden. The camera and the comparison showed who was at the desk. Each signal answers a different question, which is why the tables take them one at a time.

In the same run, Siddharth K.'s Some concerns came from page events alone: pastes from outside the page, and repeated time away.

Screen recordings and Continuous video sit on the timeline for you to watch. Recagn's rules read page events and webcam stills.

Only you see Integrity. It never enters the score, the ranking or the client's Company Report. Recording what the browser reports comes off your plate. What it means stays yours. Recagn does the screening. You make the calls.

Each check, what it records and what it adds to an interview's credits are on Recagn's AI proctoring page.

Questions people ask

Can proctoring see a second device?

Not directly. No browser feature reports a phone or a laptop beside the computer, so any tool sees one only if the webcam or a shared screen happens to show it. The surest check is the live round: ask the candidate to go one step deeper on an answer they gave.

Does a secure browser like Talview's control more than an interview page can?

Yes, on the computer it is installed on. Talview describes its Secure Browser as a one-time install for Windows and macOS that blocks messenger apps and other software, disables the cut, copy and paste shortcuts and stops the window being minimised (Talview Secure Browser, read 17 Nov 2026). A phone on the desk is a separate device, outside any browser.

Can a website tell when developer tools are open?

No browser feature reports it. A page hears key presses only while it has focus, so it can see a developer-tools keyboard shortcut, but not developer tools opened from the browser menu.

What does browser-based proctoring cost per interview?

On Recagn, activity tracking, the ID check by face match and screen recording add no credits. Webcam Snapshots add one credit to an interview, at most ₹50 on Starter before GST, and Continuous adds two for the first 60 minutes. Each check and its credits are on Recagn's AI proctoring page.

See it run on one of your roles.