Skip to main content
← Browse the editorial archive

Deepfake detection: study limits and decisions

This article is part of Certifyd’s historical security editorial archive. It was revised in October 2026 to remove unsupported claims and earlier product promises. It is not a specification of currently available Certifyd features.

Read the evaluation before the percentage

Published PNAS research examines human and machine performance on defined deepfake tasks. A benchmark result is not automatically performance on every live business call.

This article does not substantiate its earlier live-call detection percentage or invented incident narrative. Results depend on the material, participants, task and evaluation conditions.

Use controlled decisions

An organisation should not approve a payment or disclose a secret solely because a video appears genuine. Use established independent approval and contact routes and retain appropriate evidence of the decision.

Certifyd's current workforce-records scope is separate from deepfake detection. Historical meeting-verification descriptions on this site are not a promise of an available live-call authentication service.

Book a demo