AI-generated and face-swapped selfies now beat liveness and slip through onboarding. TrueSecurix scores every selfie for synthetic generation, matches it to the ID photo, and returns a fraud-risk score with the evidence behind it, in one API call.
A real victim's face mapped onto an attacker's video, the most common way liveness is defeated.
Fully synthetic people from diffusion and GAN models, with no real person behind the account.
Face matching between the live selfie and the photo on the submitted ID, to catch borrowed documents.
Post the selfie and ID to /v1/verify. You get a score, a pass / review / flag decision, and the evidence, typically in under a second. Math decides, AI advises, and uncertain cases route to review, never an auto-reject.
Read the API docs
curl https://truesecurix.com/v1/verify \
-H "X-API-Key: sk_live_..." \
-F selfie=@selfie.jpg \
-F document=@aadhaar.jpg
# => "decision": "review", "risk_score": 93Yes. It scores selfies and captured frames for AI generation and face-swapping, adding a deepfake layer on top of a liveness or video KYC flow. A deepfake injected to beat liveness is still caught by the forensic analysis.
On a held-out split the model never trained on, across 10 generators, it missed zero deepfakes and auto-rejected zero real users, with about a third of genuine selfies routed to review. Full method and dataset are on the benchmarks page.
No. AI signals only advise; a confident flag needs two independent families to agree, and blurry or low-quality images widen the review band rather than triggering a rejection. Uncertain cases route to human review, never an auto-reject.
100 verifications free, no card. Point it at your own onboarding traffic.