A delivery head calls six weeks after a joining date. The engineer your panel rated strong on a payments migration can't explain it on the project, and the client is asking who you sent. Now test last quarter's lateral hires: for how many can you show that the person who passed the final technical round is the person who joined?
Your BGV report can't answer that. Neither can the ID, certificate and employment-proof check you run before round one, the step usually called pre-interview fraud detection. Both check that the documents are genuine, and in a proxy swap they are: they belong to the person who joins. Someone else sat the interview.
SHRM India reported in July 2026 that proxy interview fraud in Indian networks has evolved into semi-organized operations, in which one skilled candidate interviews on behalf of another who later reports for the job. With notice periods often 60 to 90 days in IT services, months can pass between the interview and the day anyone notices.
What the pre-interview step can do is collect the reference: a recent photo or short video of the candidate's face that every later round, and joining day, is compared against. It protects you only if a failed match can stop the offer.
A stricter document check won't help. What's missing is a comparison: unless you build in a step, nobody checks the face on each call against a face on file. Panelists shouldn't be that step. Their attention is on the answers, and the person on the call is a real person answering well.
BGV still earns its place against a different fraud: EY India's May 2025 study of over 1 million background screenings found that 32% of IT-sector candidates submitted fake documents from companies that didn't exist or where the employer denied issuing them (EY India, May 2025).
Don't reuse the Aadhaar or PAN image from the document check. An ID photo can be years old and is printed small, and a face comparison needs a recent, clear face.
Ask instead for a short consent video, or a recent color photo against a plain background with the face filling most of the frame. Video is stronger, because a photo can be any picture the sender has. Use video where a swap would cost most. For high-volume roles, a photo collected with the application adds no extra step.
Send the request automatically at shortlist, and send it, and every interview link after it, only to the email on the application. Never route it through an agency or a forwarded WhatsApp message. If an agency forwards the file, all you know is who the agency sent.
For the most sensitive roles, a recruiter takes the reference on a short live video call, with the candidate holding their original photo ID to the camera. A person checking that ID against a live face makes it much harder for a stranger to stand in.
Settle the data questions before your CHRO asks them. You will hold the faces of many people you won't hire: every shortlisted candidate, or every applicant if you take photos at application. Decide who can open a reference and how long you keep it for people who don't join, and say in the request what it is for and who will see it.
A reference proves only that later faces match whoever recorded it. A careful proxy records the reference too, and from then on every interview matches. So who recorded it decides where the swap can show up.
In the first case, you catch it before the offer only if every round that decides the hire is compared. Leave one out, and a proxy who sits only the final technical round goes through.
In the second case, joining day is the only check left. Calls with the candidate during the notice period can't replace it, because the proxy can take those as well.
A face match answers one question: is this the same person? HirePro's 2026 candidate integrity report, as reported by PTI in October 2026, found that identity fraud fell from 65% of interview malpractice in FY23 to 28% in FY26 (Economic Times CIO (PTI), reporting HirePro's Candidate Integrity in Job Assessments and Interviews 2026 report, Oct 2026). Device-based cheating made up 20%, AI-assisted cheating 18%, reference-based assistance 15%, and camera or visibility manipulation 13% (Economic Times CIO (PTI), reporting HirePro's Candidate Integrity in Job Assessments and Interviews 2026 report, Oct 2026). That is one platform's data, from an annual sample of around 1 million assessments and 1 lakh interviews between FY21 and FY26, supplemented by industry expert input (Economic Times CIO (PTI), reporting HirePro's Candidate Integrity in Job Assessments and Interviews 2026 report, Oct 2026), so check your own mix.
Interview fraud detection that stops at a face match misses most of the malpractice in those figures. Against a second screen or a helper off camera, the interview itself does more than any identity check. Keep cameras on for the whole round. Ask fewer textbook questions and more that start from the candidate's own work: a decision on a past project, then one condition changed.
A helper can still feed answers, but each question builds on the last answer, so the pauses are easier for a panel to notice. Real-time AI help needs its own controls, covered in how agentic AI changes candidate fraud detection.
The fair objection is friction: candidates with other offers walk away from heavy checks. So match the checks to what a swap in that role would cost.
A mismatch that sits in a report changes nothing. ACFE's 2022 Report to the Nations, a global survey cited by SHRM India, found that 21% of organizations hit by employment fraud went ahead with onboarding despite red flags raised during screening (SHRM India (citing ACFE, 2022 Report to the Nations), Jul 2026). The data isn't Indian, but the pressure is familiar: with a client start date close, a flag that someone has to look up tends to lose to the deadline.
So by default, an offer to a candidate with an unresolved mismatch doesn't go out. Only a named person can override it, and they record why. Keep that authority away from the recruiter and hiring manager who own the requisition's deadline. In most teams it sits with the TA head or one delegate.
The gate also needs time. The comparison runs on the recording, so the result arrives after the call, and a Friday final round with a Monday offer may leave no room to settle a flag. Measure how long your last few flags took to resolve, and write that buffer into the hiring-manager SLA.
Comparisons fail for innocent reasons too: a poor reference, bad lighting, an odd webcam angle, a new beard. So tell the candidate only that you need to reconfirm identity before the next step, and invite them to a short camera-on call with the original ID they submitted. A recruiter who wasn't on the panel runs it, with the reference and the round's recording open. If the face on the call matches the reference but not the round's recording, someone else sat that round.
If the call clears them, record why, because a run of cleared flags means your references are weak. If it confirms a different person, close the application with a confirmed-mismatch reason, so it counts in your quarter numbers and sits on the sourcing agency's record.
At document check-in, HR opens the reference, compares it with the person in the room and records the result with their name. If HR isn't sure, don't wait for week one. Ask a final-round panelist to spend a few minutes picking up a thread from that round, such as the migration the candidate walked through. A proxy's answers don't transfer.
If the check fails, HR tells TA the same day, and system access and client allocation wait. The HR head decides, because the person has joined and this is now an employment matter. The TA head brings the evidence: the reference, the recordings and the panelist's view. The delivery head is told the seat is on hold but doesn't decide, since the person closest to the client deadline is the one most likely to let a flag through.
The hard case is a joiner who has already resigned from their last job. Don't let the resignation decide it. A wrong call would cost an innocent joiner the job they gave up, which is why the panelist conversation comes first. If that confirms someone else sat your interview, the resignation changes nothing you know: nobody on your side has assessed this person, and they were part of the swap.
The sensible default is to end the engagement under your own offer terms, with your employment lawyer involved before anyone tells the joiner. Re-interviewing them for the same seat would teach the next proxy that the worst outcome is a second interview.
If identity mismatches are a small share of your cases, put your next effort into camera rules and questions that start from the candidate's own work. If they cluster in one agency or one panel, tighten the reference and the gate there first.
By hand, this works for a few roles a month.
The reference. RippleHire's Candidate Impersonation Verification collects a consent video or a candidate photo as the reference. The consent video email goes out automatically the first time a candidate moves to Shortlisted (or an equivalent round), and a recruiter can also send it manually or upload a video on the candidate's behalf. Candidates can add a photo when they apply on the careers site.
The comparison and the gate. Shield, RippleHire's interview integrity feature, works on Google Meet and Microsoft Teams interviews in rounds set up as Shield-enabled. After the interview, it compares the recording with the reference and shows CLEAR or BREACH on the candidate's timeline. The Shield report adds a 0 to 10 integrity score alongside the face match. An offer to a candidate whose overall status is BREACH is restricted unless an authorized recruiter overrides it. Agree internally who may use that override: the TA head or one delegate, never the recruiter carrying the requisition's deadline.
The limits. Face match results need identity verification configured and reference media collected. An administrator must switch on Shield Impersonation first, and the panel must let Shield into each meeting. Shield covers only those video interviews, so phone rounds and joining day stay with your team. It doesn't analyze interviews scheduled before it was enabled, and it doesn't recommend hiring or rejecting.
If this quarter's test found a deciding round nobody compared, or an offer released with a flag still open, the gap is usually where the checks sit, not how carefully your panels look. Book a RippleHire demo and we will map the consent video request, Shield-enabled rounds and the offer restriction onto your own interview stages.
One Indian figure comes from a single vendor's platform. HirePro's 2026 candidate integrity report, as reported by PTI in October 2026, found that interview-related cases rose to 2.5% in FY26 from 1.4% in FY22 (Economic Times CIO (PTI), reporting HirePro's Candidate Integrity in Job Assessments and Interviews 2026 report, Oct 2026). The coverage doesn't say what that percentage is measured against, so read it as a direction on that platform, not a rate for your funnel. To get your own rate, compare every deciding round for one high-volume lateral role for a quarter, and count confirmed mismatches separately from cleared flags.
The dependable way is a face comparison: each deciding video round, and joining day, checked against a reference collected before round one. Without one, panels are left with softer signs. A camera that fails just as the technical round starts, a face kept half out of frame, or a candidate who can't recall what they told the previous panel are all worth noting. None of them proves a swap, and an honest candidate can have a bad connection. Log them against the round, panel and source, and look for a pattern before acting on any one sign.
Partly. AI can compare the face in a recorded video interview with the candidate's reference and report how closely they match. People are poor at this kind of judgment: a 2025 Wiley meta-analysis, cited by SHRM India, found that untrained people identified deepfake content only 55.54% of the time, slightly better than chance (SHRM India (citing a 2025 Wiley meta-analysis), Jul 2026). That finding covers deepfake content in general, not interviews, but it suggests a panelist's eye is not an identity check. AI can't judge a round it never saw, such as a phone call, and it shouldn't make the hiring decision.
Three, before the first round: confirm the documents are genuine, get a face reference on file, and make sure interview links go only to the email on the application. SHRM India's July 2026 guidance makes the same timing point, describing screening validation as verifying identities and credentials "before the interview pipeline, not after the offer" (SHRM India, Jul 2026). Run the full document check at shortlist rather than at application, so you hold ID documents only for people you will actually interview.