How do I know if a candidate is using a deepfake in a video interview?

Watch for mismatches. Lip movement that lags the audio, eyes that never track naturally, lighting that stays flat when the person moves, and a face that blurs at the edges during fast motion. Ask the candidate to turn their head fully sideways or hold a hand near their face. Most real-time deepfake tools break on those moves. Pair that with a live, unscripted question the candidate could not prep an AI to answer.

Who is liable if we place a fraudulent candidate at a client?

You are, in most contracts. If a placed worker turns out to be a fake identity or a stand-in, your firm carries the compliance and reputational hit, plus any clawback on the fee. That is why a documented fraud-detection protocol matters. It shows the client you did reasonable screening, and it gives you a consistent process your recruiters actually follow instead of guessing case by case.

Do we need to buy fraud-detection software to handle this?

Not to start. Most of the first wins are process, not tech. Live identity checks, structured verification steps, and a trained eye on video interviews cost nothing but discipline. Tools help you scale once the process works. Buying software before you have a protocol just automates a broken workflow, so fix the protocol first.

Deepfake Candidates: Fix Your Screening Before Year-End

Candidate fraud is no longer rare. With 72% of staffing buyers reporting it and Gartner projecting 1-in-4 profiles will be fake by 2028, firms placing remote tech roles need a documented fraud-detection protocol. Here is what to add before year-end.

Fake candidates are now a normal part of your pipeline, whether you see them or not. This is not a future problem. It is a this-quarter problem, and most staffing firms have no written plan for it.

Staffing Industry Analysts reports that 72% of staffing buyers say they have encountered candidate fraud, in a report published October 8, 2026. That is not a fringe group. That is three out of four buyers telling you they have been burned.

And it gets worse from there. Staffing Industry Analysts notes that Gartner predicts one in four job candidate profiles globally will be fake by 2028. The same piece points out that Experian rates deepfakes outsmarting HR as the second-highest fraud threat of 2026.

If you place remote IT or tech workers, you are the prime target. Those roles are high-value, fully remote, and easy to fake your way into with a borrowed resume and a working camera trick. The person who shows up to the interview is not always the person who shows up to the job.

Why this hits staffing firms harder than direct employers

A direct employer screens one pipeline for one job. You screen dozens of pipelines a week and move fast because speed is the product. That speed is exactly what fraud rings count on.

Here is the operational gap. When a fake candidate slips through, the liability lands on you first. Your contract with the client almost always puts the screening burden on your firm. If the placed worker turns out to be a stand-in or a stolen identity, you eat the clawback, the cleanup, and the damaged relationship.

Most firms have no documented fraud-detection protocol. They have a gut feeling and a busy recruiter. That worked when fraud was rare. It does not work when a quarter of profiles are fake.

This is a scaling problem, not just a security problem. As you grow, you cannot rely on your two best recruiters to spot a deepfake by instinct. You need a process every recruiter runs the same way, every time, so the detection does not fall apart the moment you hire your next ten people.

What to add to your screening playbook before year-end

You do not need to buy software this week. You need a written protocol your team actually follows. Start with these.

1. A live identity verification step you control

Require a live, on-camera moment where the candidate holds up a government ID next to their face. Then ask them to do something a deepfake struggles with in real time. Turn the head fully sideways. Hold a hand flat across part of the face. Say a random phrase you give them on the spot.

Real-time deepfake tools often break on sudden movement or occlusion. The face blurs, the edges warp, or the lip sync falls apart. Train your recruiters to watch for it and to end the call if something feels off.

2. One unscripted technical question per interview

Fraud rings prep candidates with scripted answers and AI copilots feeding them lines. Beat it with specificity. Ask about a real decision from a project on their resume. Ask why they chose one approach over another. Ask them to screen share and walk through something live.

A real engineer talks about tradeoffs and mistakes. A fake reads a clean, generic answer with a two-second delay while the AI catches up. That delay is your tell.

3. A documented verification trail on every placement

Write down what you checked and when. ID verified on camera on this date. Reference called at this number. Live technical screen completed by this recruiter. This record protects you if a client ever comes back on a bad placement, and it forces consistency across your desk.

If it is not written down, it did not happen as far as your client or your insurer is concerned.

4. A reference call you actually dial yourself

Fraud rings supply fake references on supplied numbers. Do not call the number on the sheet. Find the reference independently, through the company directory or a verified profile, and call that. It takes five extra minutes and closes a door the good scams rely on.

Where AI helps and where it hurts

AI is on both sides of this fight. The same tools that generate deepfakes can help you catch them, but only if your process is clean first.

AI screening tools can flag interview behavior that points to cheating, score video for signs of manipulation, and cross-check identity data at scale. That is real value once you are placing enough volume that manual review cannot keep up.

But here is the trap. If your underlying process is sloppy, AI just automates the sloppiness. It flags a hundred candidates a day and your team ignores the alerts because nobody owns the follow-up. Buy the tool after the protocol works, not before. The protocol is what makes the tool pay back.

This is the same lesson that shows up everywhere AI touches your firm. Clean process and clean data first. Then automate. Skip that order and you spend money speeding up a broken workflow.

The cost of doing nothing

A single bad placement in a remote IT role can run you the fee clawback, the replacement cost, the client's lost trust, and in a regulated industry, a compliance exposure you do not want to explain. Multiply that across a year where a growing share of profiles are fake.

The firms that handle this well are not the ones with the fanciest software. They are the ones with a written protocol every recruiter follows, so detection does not depend on who happens to run the call.

Your next step this week

Pick one requisition type you fill most often, likely a remote tech role. Write a one-page fraud-detection protocol for it. Include the live ID check, the unscripted question, the independent reference call, and the verification trail you will log. Walk your recruiters through it in one meeting this week and require it on the next ten placements.

One page. One role. This week. That is how you close the gap before year-end, before the fake candidate closes it for you.

The Deepfake Candidate Crisis: What to Add to Your Screening Playbook Before Year-End