Deepfake Detection

Methods that detect synthetic or manipulated media (audio, images, video) used to impersonate people in identity verification and biometric systems.

Overview

Deepfake detection aims to identify AI-generated or heavily manipulated media intended to defeat identity verification (e.g., selfie checks) or to enable fraud via impersonation.

How it’s used in identity systems

  • As a signal alongside PAD/liveness, device checks, and document verification.
  • To flag suspicious submissions for step-up or manual review.
  • To harden enrollment and re-verification, where synthetic media can create or take over accounts.
  • To protect consumer platforms from fake profiles, account farms, and impersonation.
  • To reduce onboarding risk in payments, crypto, and other regulated financial services.

Common challenges

  • Rapidly evolving generation methods and attack techniques.
  • Domain shifts (lighting, cameras, compression) that can affect detector performance.
  • Distinguishing synthetic-media detection from biometric liveness. In practice, the two are often combined, but they are not identical controls.

References

Vendors using Deepfake Detection

Latest Data Cards

  • Data Card

    Socure takes growth investment at $5.2B valuation and buys Fravity

    2026-08-28CC-BY-4.0digital-iddeepfake-detectionsocure

    Socure raised a $156 million growth round at a $5.2 billion valuation and acquired agentic fraud investigation platform Fravity, integrating it as RiskOS_Agents.

    • The $156 million round was led by Summit Partners with Goldman Sachs Alternatives and Wells Fargo.
    • Socure reported $364 million in annual recurring revenue.
    • Fravity automates fraud and anti-money laundering alert resolution as RiskOS_Agents.
  • Data Card

    EU AI Act transparency rules for deepfakes and biometric systems take effect

    2026-07-31CC-BY-4.0deepfake-detectionfacial-recognition

    Article 50 of the EU AI Act took effect on August 2, 2026, requiring disclosure of AI interactions, machine-readable marking of generated content, and notification of biometric categorisation.

    • Generated content must carry machine-readable markings.
    • Deployers must notify individuals exposed to emotion-recognition or biometric-categorisation systems.
    • Non-compliance can draw fines up to €15 million or 3 percent of global annual turnover.
  • Data Card

    US House panel hears that the federal identity model cannot keep up with AI fraud

    2026-07-17CC-BY-4.0deepfake-detectiondigital-idsocure

    Testimony to the House Subcommittee on Government Operations argued that generative AI fraud is outpacing the identity proofing used across state and federal service portals.

    • Socure's Jordan Burris described one ring creating nearly 25,000 synthetic identities and launching more than 35,000 attacks in 30 days.
    • SentiLink's David Maimon described criminals combining stolen identities with AI-generated faces and deepfake video to defeat liveness checks.
    • Witnesses said readily available face-swapping software lowers the skill needed to mount these attacks.