Deepfake Detection evaluation questionnaire

This evaluation is stored in your browser only. We cannot see it, and it is not tied to any account. Save it to a link or create an account to keep it across devices — you can export it at any time either way.

All 12 vendors in this category.

• baselineRFIWhat media types does the platform detect (voice/audio, video, still images, or all three), and what is the measured detection accuracy against current-generation generative models specifically — not older, easier-to-detect deepfakes from a year or more ago?weight 3
  Answer key — what a strong answer shows

  This field moves extremely fast; a vendor citing accuracy figures against outdated deepfake-generation techniques is not evidence of current effectiveness. Insist on recency of the test set used to measure accuracy.

• baselineRFPDescribe real-time detection capability for live scenarios (e.g., a live video call, an inbound phone call) versus post-hoc analysis of already-recorded media, and what latency is added to a live call if real-time detection is claimed.weight 3
  Answer key — what a strong answer shows

  Real-time detection during a live call (e.g., CEO-fraud voice-cloning scenarios) is a materially harder and more valuable capability than post-hoc file analysis; ask for a measured latency figure if real-time is claimed, since added delay affects usability.

• baselineRFIHow does the platform keep pace with new generative-AI model releases that may produce deepfakes the current detection model hasn't been trained to catch — what is the update cadence for detection models, and is there a documented lag between a new generation technique appearing and detection catching up?weight 3
  Answer key — what a strong answer shows

  A credible vendor is honest about the fundamental cat-and-mouse dynamic in this category and describes a real update cadence, rather than implying permanent, static effectiveness against all future generation techniques.

• baselineRFIWhat is the false-positive rate — how often does genuine, unaltered media get flagged as a deepfake — and what is the business impact of a false positive in the platform's primary use case (e.g., blocking a legitimate executive's voice instruction)?weight 3
  Answer key — what a strong answer shows

  False positives in this category carry real business disruption risk (blocking a genuine urgent instruction); a vendor should state a measured false-positive rate and describe how a flagged-but-legitimate case gets resolved quickly.

• baselineRFPDetail the primary use case this platform targets (voice-authentication/CEO-fraud prevention, video content moderation, identity-verification liveness checks) and provide a production customer reference specific to that use case, since deepfake detection needs differ substantially by scenario.weight 3
  Answer key — what a strong answer shows

  This category spans very different use cases with different technical approaches; a vendor's general-purpose claim should be grounded in a specific, relevant production reference rather than accepted at face value.

• baselineRFIHow does the platform integrate into existing workflows where deepfake risk matters most — e.g., flagging a suspicious voice instruction within a wire-transfer approval workflow, or flagging suspicious video within a KYC identity-verification pipeline — versus operating as a standalone analysis tool disconnected from the actual business process at risk?weight 3
  Answer key — what a strong answer shows

  Detection with no workflow integration point provides limited real protection; ask specifically how a detection actually intercepts or flags the at-risk business process in practice, not just in a demo.

• baselineRFPExplain what evidentiary/forensic output the platform provides when it flags content as likely synthetic — is the explanation interpretable (specific artifacts identified) or a black-box confidence score with no supporting detail useful for escalation or law-enforcement referral?weight 3
  Answer key — what a strong answer shows

  An interpretable explanation (specific detected artifacts) is more useful for escalation decisions and any downstream legal/forensic process than an opaque confidence score alone.

• baselineRFIWhat is the pricing model — per media item analyzed, per call/session monitored in real time, or a flat enterprise tier — and what is the cost impact of enabling real-time detection on every call versus periodic spot-checking?weight 3
  Answer key — what a strong answer shows

  Look for transparent pricing that reflects the real cost difference between blanket real-time monitoring and selective/spot-check analysis, since always-on real-time detection is materially more resource-intensive.

• baselineRFPOnce content is flagged as likely synthetic, what is the actual response workflow — automatic blocking, mandatory human review before any action, or just a logged alert requiring the business process to separately notice and act — and what is a customer-referenced example of this workflow stopping a real fraud attempt?weight 3
  Answer key — what a strong answer shows

  Strong answers describe a real, tested response workflow with a concrete fraud-prevention example, not just confirmation that detection and flagging occur.

• baselineRFIAre legal/compliance considerations addressed for using deepfake-detection findings in a formal proceeding (HR investigation, legal action) — is the detection output and its underlying methodology defensible/admissible, and has this been tested in a real case?weight 3
  Answer key — what a strong answer shows

  Given detection findings could be used to justify serious action (termination, legal claims), ask for a specific answer on evidentiary defensibility, not just detection accuracy in isolation.

• baselineRFIDetail historical trend reporting on deepfake-attempt volume and sophistication trends observed by the platform over time, suitable for demonstrating threat-landscape awareness to leadership.weight 3
  Answer key — what a strong answer shows

  Trend-over-time reporting is a distinct capability from individual detection events — confirm this exists as a maintained, exportable report.

• baselineRFIFor voice-based detection specifically, is accuracy consistent across different languages, accents, and audio quality conditions, or does detection skew toward higher accuracy for specific languages/accents the model was primarily trained on?weight 3
  Answer key — what a strong answer shows

  Voice-deepfake detection models can have real accuracy disparities across languages and accents — ask for an honest, specific answer rather than an unqualified accuracy claim that may reflect only the primary training-data demographic.

• baselineRFPHow does this platform relate to the customer's existing identity-verification/KYC tooling — is deepfake detection a genuinely distinct, complementary layer, or does it substantially duplicate liveness-detection capability an identity-verification platform already provides?weight 3
  Answer key — what a strong answer shows

  This is a real, common buyer question given the functional overlap with identity-verification's own liveness-detection capabilities — a vendor should give an honest answer about the boundary and complementarity.

• baselineRFIHas the detector itself been red-teamed or adversarially tested against state-of-the-art evasion techniques specifically designed to fool detection models, rather than only validated against older, easier-to-detect deepfakes?weight 3
  Answer key — what a strong answer shows

  Detection accuracy against yesterday's deepfakes says little about resilience against deepfakes specifically crafted to evade this detector — ask for evidence of genuine adversarial red-teaming, not just a static accuracy benchmark.

• baselineRFPWhat is a customer-referenced onboarding timeline from contract signature to the platform providing genuinely reliable detection integrated into the specific at-risk business process (e.g., wire-transfer approval, KYC pipeline), and what customer-side workflow-integration effort does that require?weight 3
  Answer key — what a strong answer shows

  Strong answers give a concrete, customer-validated timeline for actual business-process integration, not just standalone tool availability, and are honest about the customer-side effort required.

These questions and weights are the reviewed standard for this category — start scoring straight away. To change them, or to import your own questions from a spreadsheet and reuse them on every evaluation, create a free buyer account.