HR + recruiting teams

Spot candidate fraud before the offer goes out

Resemble analyzes audio and video during interviews, alerting recruiters to suspicious activity while the call is still underway. It also checks whether the same person appears across interview rounds and gives your team evidence to review before making a hiring decision.

Contact us
CASE STUDY

A recruitment automation platform runs over 250,000 candidate screening calls a month and scores every candidate afterwards. At that volume nobody is listening for a cloned voice, and nothing in the stack is checking for one today.

250,000
candidate screening calls a month, on one platform

How it works

Three checks, at the three points in hiring where a candidate can be someone other than who they say. Nothing changes about how the interview is booked or run.
Before the call: The candidate's ID goes through the API and comes back with a verdict and the altered region, if there is one.
During the interview: Audio, video and identity are checked while the call runs.
Across the rounds: The candidates identity from the first interview is enrolled, so a different face or voice in a later round is flagged.
View Docs

How fake candidates get through a video interview

Candidate substitution

A proxy interview happens when someone other than the applicant takes part in an interview, often sitting the technical round before the actual candidate starts the job. This kind of candidate fraud can go unnoticed when interviewers don't compare identities across rounds.

Face swaps on the live call

Free tools can now put a different face on a live video call through a virtual camera. A convincing deepfake interview can make it difficult for recruiters to know who's actually on screen, even when the person looks and sounds like the candidate.

Cloned voices on the phone screen

Phone screens are often less closely monitored than video interviews, making them a target for cloned voices and automated calls. Voice cloning detection analyzes the audio for signs of synthetic speech, helping recruiters flag suspicious calls before moving a candidate forward.

Documents at onboarding

Candidate fraud can continue after an offer is accepted. IDs, work authorization documents and professional certifications may be generated or altered to support a false identity or set of credentials. Standard checks can confirm a document's format and expiry date without establishing whether it's genuine.

What your hiring process checks, and what it misses

Your existing screening process does a good job of establishing whether someone has the background they claim. The harder question is whether the face and voice in the interview belong to the person who applied.
The Question Your Current Process Resemble
Does this candidate have the background they claim? Background checks, reference calls and work authorization verification. These checks do this job well. Nothing to add here. This part of the process already works.
Is the person in this interview the person who applied? Not assessed. Your ATS authenticates an account and a calendar invite, but nothing checks the face and voice that join the call. Audio and video are checked while the interview runs, and a synthetic participant can be flagged for the recruiter.
Is this the same candidate who sat the first round? Usually left to whoever ran the earlier interview, and to their memory weeks later. Identity matching across sessions can flag a different face or voice in a later round.
Was this ID or certificate generated? Documents are checked for format and expiry, then generally treated as genuine. Pixel-level detection can identify generated and altered documents and locate the region that was edited.
Can you show why a candidate was rejected? A recruiter's note based on their impression of the interview. That can be difficult to substantiate if the decision is challenged. A timestamped report records the evidence behind the flag, giving your team something concrete to review and explain.

What Resemble AI sends back

One API call returns the detection and the report together
Result and risk level
Deepfake or authentic, with a confidence score and a risk level.
Liveness
Whether a real person was physically in front of a camera, or whether the feed was injected from somewhere else.
Abnormalities
The specific artifacts that triggered the flag, such as distorted features, lighting that does not match, over-smoothed skin, or gibberish text on a document.

5 questions to take to your next fraud review

Most talent teams can describe their screening process in detail and have never been asked how they would know if a candidate was not real. Hiring fraud exposes this gap, and these five questions surface it quickly.
If someone joined an interview tomorrow wearing another candidate's face, what in your process would catch it?
Who confirms that the person in the final round is the same person who joined the first?
What does a recruiter do when an interview feels wrong, and who do they tell?
How many of your screening calls are voice-only, and what verifies the voice on them?
If you turned a candidate down for suspected fraud, what evidence could you show to explain the decision?

Where it runs, and what happens to your data

Where it runs
Cloud, your own VPC, or fully on-prem. Air-gapped deployments are already in production with government agencies.
What we keep
In memory processing is available, so a file is analyzed without being stored. Everything is encrypted at rest and in transit.
Certification
ISO/IEC 27001:2022 certified, and GDPR and HIPAA compliant. Full detail is in our trust center.
99.5%
Audio acccuracy on the Podonos benchmark, externally validated
<1%
False positives and false negatives, Podonos benchmark
54
Languages covered on audio from Spanish to Arabic
Frequently asked questions
Does a bot have to join the interview?
Resemble analyzes the characteristics of a voice to determine whether it was AI-generated, rather than analyzing what the speaker says. It returns a deepfake confidence score in under 300 milliseconds, without requiring transcription or recording. Detection works across 54 validated languages, and because it doesn't require recording the conversation, it can support deployments where recording consent is a concern.
What do we have to tell the candidate?
Your employment counsel should advise on this, because disclosure requirements vary by jurisdiction. Most teams tell candidates about detection in the same place they already disclose interview recording.

A privacy mode that analyzes the call without storing it can also help make the conversation about candidate data and retention more straightforward.
What happens when it flags a real candidate?
Video quality affects detection, and interviews are often where recording quality is at its worst. Compression, frame rate, poor lighting, a weak connection and the beauty filters built into conferencing tools can all contribute to a false positive.

That is why an alert goes to the recruiter rather than ending the interview. Your team stays in control of the decision, and we can calibrate the system on your own recordings before anyone commits to a performance figure.
Can the ID be checked before the interview?
Yes. A document can be submitted through the same API before the call. This is how candidate identity verification is often set up: the document is checked first, and the face on the ID can then be compared with the face that joins the interview.

For teams running high-volume screening, checking the document in advance means the interview itself can focus on confirming the identity and assessing the candidate.
Does it integrate with our ATS?
Detection runs through the API or inside the meeting platform, so it can fit into the tools your team already uses. Results can be written back into your own systems.

Native integrations with applicant tracking systems are still in development but talk to our team to learn more.
Does this catch North Korean IT worker schemes?
It can detect synthetic elements used in these schemes, including altered identity documents, a borrowed or generated face in an interview, and a voice that does not belong to the applicant.
What languages does it cover?
Audio detection covers 54 languages. The check analyzes the acoustic signal rather than relying on a transcript, so it does not need to understand what was said. This is useful when you are hiring across markets and your recruiters do not speak every language used in an interview.
Know what's real — and what's a real threat.
Join thousands of developers and enterprises detecting AI fraud and protecting their content with Resemble AI
Contact us