Detect interview deepfakes and candidate swapping on live video calls.
Deepfake detection built for the interview setting that analyzes audio and video together -add deepfake detection capabilities to your platform with a single API key.
Trusted by leading companies.
We’ve built the best deepfake detection model for the zoom call.
Off-the-shelf models weren’t built for this.
Deepfake attacks in the interview and video setting have increased 100% in the last year. We tested dozens of existing models to build on top of, but none performed well enough — they’re built for other use cases like music, call centres, and general media.
So we built one only for the interview.
The same deepfake that can prove a politician is being deepfaked isn’t the one that will protect an enterprise from recruiting threats. So we focused entirely on the interview and zoom setting — and it’s why we outperform every other deepfake detection model on the market for this use case.
Built to secure every interview.
Deepfake detection, candidate-swap checks, and enterprise-grade security — in a single API built for live video calls.
Accuracy
Purpose built for the interview setting.
Industry leading accuracy
92%+ deepfake detection accuracy, highest among deepfake detection vendors for interviews and video calls.
Trained on live data from real attacks
Tofu models are trained on proprietary, live interview data from genuine candidates and bad actors using the latest attack techniques.
Synthetic audio and video analyzed in a single call
Interviews are analyzed with audio and video combined using data from both inputs to identify deepfakes.
Candidate swapping
Detect candidate swapping, proxy interviewers and cheating
Ensure the same candidate shows up to each call
Notify your customers when someone different shows up to interview on behalf of the original candidate.
Even when their camera’s are off
Analyzes video and audio to catch proxies who have their video cameras on or off.
Catch AI-assisted interview cheating
Detects behavioural patterns highly associated with cheating.
Security & scale
Enterprise security and scale
Zero data retention
No video or audio is stored by default. Retention is configurable if you need it -- otherwise nothing is retained after the response is returned.
Not trained on your data
Interview footage is used to produce a result and nothing else. We don't use customer data to train, fine-tune, or improve our models.
Built for scale
99.99% uptime. Weekly model updates without changing the response format your systems already integrate against.
Built to secure every interview.
Deepfake detection, candidate-swap checks, and enterprise-grade security — in a single API built for live video calls.
Book a demoAccuracy
Purpose built for the interview setting.
Industry leading accuracy
92%+ deepfake detection accuracy, highest among deepfake detection vendors for interviews and video calls.
Trained on live data from real attacks
Tofu models are trained on proprietary, live interview data from genuine candidates and bad actors using the latest attack techniques.
Synthetic audio and video analyzed in a single call
Interviews are analyzed with audio and video combined using data from both inputs to identify deepfakes.
Alex Morgan
Software Engineer
Unknown individual
Identity mismatch
Candidate swapping
Detect candidate swapping, proxy interviewers and cheating
Ensure the same candidate shows up to each call
Notify your customers when someone different shows up to interview on behalf of the original candidate.
Even when their camera’s are off
Analyzes video and audio to catch proxies who have their video cameras on or off.
Catch AI-assisted interview cheating
Detects behavioural patterns highly associated with cheating.
Security & scale
Enterprise security and scale
Zero data retention
No video or audio is stored by default. Retention is configurable if you need it -- otherwise nothing is retained after the response is returned.
Not trained on your data
Interview footage is used to produce a result and nothing else. We don't use customer data to train, fine-tune, or improve our models.
Built for scale
99.99% uptime. Weekly model updates without changing the response format your systems already integrate against.
More from tofu.
Insights, research, and updates on applicant fraud.
IT Worker Fraud Has an Identity Problem
We pulled every case our resume fraud detection tool flagged as a possible impersonation. What we found — deepfakes, stolen LinkedIn photos, and real-time face-swapping software — was worse than we expected.
Resume Fraud Detection Needs A Human Score
We've spent two years building one of the largest fraudulent candidate datasets in the industry — over 5M applicants, billions of data points. Here's the signal that surprised us most.
tofu. + Gem: Applicant Fraud Detection in Gem's Product Suite
Recruiting has quietly entered a crisis. The combination of generative AI and remote work has made it incredibly easy for bad actors to flood hiring pipelines with fake identities, deepfakes, and AI-generated candidates.