Forensic-grade deepfake analysis
Cyveck classifies suspected manipulated media with per-frame confidence tiers, an analyst-reviewable explainability layer, and a tamper-evident audit log.
Request a demo →Analyst workflow
An analyst submits a video or image. Cyveck detects faces, classifies each detection against a model trained on a diverse corpus of authentic and manipulated face media, and routes the result into one of four confidence tiers.
Trust
High margin, stable under perturbation. Verdict accepted without analyst review.
Provisional trust
High margin, unstable under perturbation. Delivered with a spot-check flag.
Priority review
High margin but localized sensitivity. Full analyst review required; explainability heatmap shown.
Full review
Low margin, high variance. The model is explicitly inconclusive. Analyst decides.
Each case produces a PDF report with the verdict breakdown and an evidentiary audit log with cryptographically chained entries.
Deliverables
Every case produces two artifacts.
A case report PDF
Overall verdict, per-frame breakdown, explainability heatmap for the highest-scoring manipulated frame, and provenance metadata. Available to pilots.
An audit log
A structured log with cryptographically chained entries covering every lifecycle event from case submission to verdict. Tampering with any entry breaks the chain and is detectable.