Can candidates cheat on one-way video interviews?

Yes, candidates can cheat on one-way video interviews, and more easily than on a live call. A one-way video interview has nobody on the other end, so nobody can ask a follow-up or notice that the answer was read off a second screen. A stand-in can record it, and a deepfake can be rendered ahead of time instead of live. Keep the format for volume, then add one short live step before any offer.

Most of what's written about deepfake interviews (our own posts included) pictures a live call: an interviewer, a candidate, and a face that goes soft at the edges when it turns. The checks people recommend all need someone on the call to ask for them, and in a recorded interview there's nobody there.

What is a one-way video interview?

A one-way video interview (also called an asynchronous or recorded interview) is basically a screen with nobody on the other end: the candidate gets a link and records answers to on-screen questions on their own time. Nobody from the hiring team is watching while it happens. Recruiters review the videos later, or software scores them, or both.

Teams use them for a good reason. When a role gets 2,000 applicants, nobody has time for 200 phone screens, and a recording you can watch at 1.5x speed is a reasonable trade.

Why is fraud easier in one-way video interviews?

One-way interviews are easier to fake because everything that catches a fake on a live call needs a person on the other end, and four of those checks disappear in a recording.

The first is that nobody asks a follow-up. On a live call, a scripted answer runs into "Walk me through how you'd debug that" or "Oh, you're in Austin? What part?" The FBI recommends those soft questions about location and education in its January 2025 warning about North Korean IT workers, the same notice that says those workers have been seen using AI and face-swapping technology. A recorded question gets the same answer whether it's true or not.

Nobody asks for a movement either. When Palo Alto Networks' Unit 42, in April 2025, built a real-time deepfake for job interviews in about 70 minutes, the weak spots were fast head turns, a hand crossing the face, sudden lighting changes, and lips drifting out of sync with the voice. Each one is found by asking the candidate to do something. In a recording, the candidate does only what they planned to do.

Time also stops being a constraint. Unit 42 noted that a better-looking fake needed more time and a stronger graphics card. A live deepfake has to render in real time. A recorded one can take all night, and the fraudster can watch it back before submitting. (Unit 42 only tested live fakes, so that part is my inference.) If your platform allows retakes, a bad take gets recorded again.

And the screen is free. A candidate on a live call who keeps glancing down gets asked about it, or at least feels watched. On a recording, an AI tool can draft the answer while the question is still on screen, and the candidate reads it aloud. A stand-in can do the whole thing, which is the recorded version of seat swapping.

Doesn't research say people fake less in one-way interviews?

It does, for one kind of faking. A 2023 field study of 144 interviewee and interviewer pairs, by Hung-Yue Suen and Kuo-En Hung at National Taiwan Normal University, compared deception in asynchronous and live video interviews. Applicants in the recorded format showed less deceptive self-presentation on every dimension measured, and AI-assisted scoring cut down on making things up in both formats. Interviewers struggled to spot the deception either way, except for heavy faking in the recorded ones.

I believe that result. But the study measured real applicants polishing their own stories: overstating a skill, smoothing over a gap. It didn't look at whether the person on the tape was the applicant at all. Identity fraud is a different problem, and a stand-in with a well-rehearsed answer may look more honest than a nervous real candidate.

What still shows up in a recorded interview?

Less than a live call gives you, but recruiters who watch a lot of these learn to spot a few things. The ones I'd look for:

• Eyes that track text: steady left-to-right movement while answering, then a look up at the end of each line. People answering from memory tend to look away while they think.
• A pause, then perfect fluency. A few seconds of silence after every question, followed by a polished answer with no false starts. Real people start talking and fix the sentence halfway through.
• Lips slightly out of time with the voice. You'll catch it more easily on a recording than live, because you can scrub back and watch the same three seconds again.
• Edges that don't behave: hairline and glasses frames that blur or shimmer when the head moves, or lighting on the face that doesn't match the room.
• A different person later. The face or the accent on the recording doesn't match the person who shows up to the first live round. This one is the most useful, and it only works if you compare.

None of these proves anything alone (why deepfake detection needs multiple signals goes into that). Cheap webcams lag, and plenty of honest candidates bring notes to a one-way interview because nobody told them not to. Treat a sign as a reason to look closer at the next stage. What to do when you suspect a deepfake covers how to handle it without accusing a real candidate.

What should TA teams change?

Keep the one-way interview. It does its job, which is cutting a huge applicant pool down to a shortlist. Then make sure nobody gets an offer on recordings alone.

1. Add one short live step before the offer: 15 minutes, camera on, with a person from the hiring team. If five finalists reach that stage per hire, that's 75 minutes of recruiter time per hire. Compare that with what it costs to onboard someone who isn't who they said they were.
2. Compare the live person to the recording. Same person as the tape? Have the recruiter who watched the recorded answers run the live call, or pull up a clip beforehand.
3. Ask the soft questions the recording couldn't, like where they live and where they studied. A real candidate answers without thinking.
4. Write down which stages were recorded. If something goes wrong later, you'll want to know whether anyone from your team ever saw this person live.

The FBI's advice in the same notice is to do as much of hiring and onboarding in person as you can. Most remote teams can't fly every finalist in, and a live camera-on step is the realistic minimum.

But Joseph, we use one-way interviews because we don't have time for live screens. Fair, and I'm not suggesting you go back to screening everyone live. The live step goes at the end, for the handful of people you're about to hire, where the cost of being wrong is highest and the number of calls is smallest.

Where does deepfake detection fit?

Deepfake detection fits at the live step: the 15-minute call before the offer, and any live round before it. Tofu makes candidate fraud detection for hiring and security teams, and our deepfake detection runs in the background of interviews on Zoom, Google Meet, Teams, and AI note-takers like BrightHire. It looks at lip syncing, eye movement, facial construction, voice patterns, and hints of AI overlay, and it checks that the candidate on this call is the same person from the last one. Nobody has to ask a finalist to wave a hand in front of their face.