Sensity AI deepfake detector

Vendor Sensity B.V. Official site sensity.ai/deepfake-detection-for-video-i... How this card is sourced

AIDI-D-0038

Updated ·

Sensity AI accepts files of up to 500 MB and audio or video of up to one hour through its cloud API, per the vendor's documentation (docs.sensity.ai, read 9 November 2026). Sensity AI is a deepfake detector for video, images and audio, not text: every submission runs through four services, "Face Manipulation, AI-Generated Content Detection, Voice Analysis, and File Analysis", and returns a binary verdict of suspicious or valid with confidence scores and heatmaps (docs.sensity.ai, read 9 November 2026). The vendor is Sensity B.V. of Amsterdam, which dates its founding to late 2018 and sells to government, law enforcement and forensic buyers; it publishes no prices and holds new cloud sign-ups for manual approval (sensity.ai/about-sensity-ai and the documentation, read 9 November 2026). The card joins the detector index beside the other media cards, and how this card is sourced explains why a tool with no study record and no price list still gets the same fourteen fields.

What Sensity AI detects

The Sensity register entry: field, value, source and read date
FieldValueSource
Content types detectedVideo, image, voice, file forensicssensity.ai/deepfake-detection-for-video-i...
Generators coveredFace manipulation: AI face swap, AI lip sync, AI face reenactment, AI face morphing, AI-generated content: AI avatars, synthetic GAN faces, text-to-image, text-to-video, Audio: synthetic voice and AI voice cloning, The tech page shows a Model Generator attribution (example: Flux) beside the AI-generated content verdict; no generator list is publishedsensity.ai/deepfake-detection-for-video-i...
Input limitsCloud API and platform (documentation, read 9 November 2026): files up to 500 MB by direct upload or public URL; audio and video up to 1 hour; video up to 2560 px on the longest dimension; images jpg, jpeg, png, tiff, gif, webp, jfif; video mkv, flv, mov, mp4, avi, webm, 3gpp, m4v; audio wav, mp3, m4a, ogg, aac, flac, pcm; URLs from X, Instagram, TikTok, Facebook, LinkedIn, Reddit, Pinterest, YouTube, Vimeo, Dailymotion and Twitch; submissions rate-limited to 100 requests a minute with a workspace quota per billing period (monthly, quarterly or yearly). On-premise deployments do not enforce fixed media limits. Every submission runs four services, Face Manipulation, AI-Generated Content Detection, Voice Analysis and File Analysis, and returns a binary verdict, suspicious or valid, with confidence scores and heatmaps.docs.sensity.ai/resources/formats-limits
Free tierNone published. The site offers Get started and Start a trial buttons and a Talk to sales form; new cloud accounts are held for manual review and cannot log in or create API keys until approved (documentation). No prices, quotas or trial terms are publishedsensity.ai
APIYes, docs.sensity.aidocs.sensity.ai
IntegrationsREST API v2 (api.sensity.ai) with signed upload URLs, webhooks and CSV export, SDK, Web app at platform.sensity.ai with teams, roles and an analytics dashboard, URL submission from X, Instagram, TikTok, Facebook, LinkedIn, Reddit, Pinterest, YouTube, Vimeo, Dailymotion and Twitch, On-premise server, workstation and USB dongle deployments, Integration at evidence ingestion inside forensic environments (home page)docs.sensity.ai
Data retentionPrivacy Policy (last updated January 2024, Sensity B.V.): We store the personal information we receive as described in this Privacy Notice for as long as you use our Services or as necessary to fulfill the purpose(s) for which it was collected, provide our Services, resolve disputes, establish legal defenses, conduct audits, pursue legitimate business purposes, enforce our agreements, and comply with applicable laws. We will delete your biometric data within three years of your last interaction with the Services. The policy describes selfie videos and identification documents as the source of biometric data and states no separate period for submitted media files. Documentation: submissions can be deleted through the API (DELETE on submissions, with a 409 while processing).sensity.ai/privacy-policy
Launch year2018sensity.ai/about-sensity-ai
OwnershipSensity B.V., Javastraat 25 I, 1094GZ Amsterdam, Netherlands (Terms of Service, last updated October 2026, governed by Netherlands law with binding individual arbitration; Privacy Policy, last updated January 2024). The about page (dated 3 February 2026) names co-founders Giorgio Patrini, a former postdoctoral researcher at the University of Amsterdam, and Francesco Cavalli, and says that late in 2018, Sensity AI was born as the world's first deepfake detection company. Sold to government, law enforcement and forensic buyers; cloud sign-ups are held for manual approval by a Sensity administrator (documentation).sensity.ai/terms-and-conditions

The product page lists four targets: "Face Manipulations", "Synthetic Voices", "AI Generated Videos" and "AI Generated Images" (sensity.ai/deepfake-detection-for-video-image-audio, read 9 November 2026). Under face manipulation it names AI face swap, AI lip sync, AI face reenactment and AI face morphing; under AI-generated content, AI avatars, synthetic GAN faces, text-to-image and text-to-video; under audio, synthetic voice and AI voice cloning. The tech page adds that the face model "analyzes content frame-by-frame, detecting altered faces with a binary verdict, confidence score, and manipulation type", with a red bounding box and pixel-level heatmaps, and shows a "Model Generator" attribution in its example result (sensity.ai/tech-stack, read 9 November 2026). No list of generators detected is published, so the generators field records the manipulation types instead.

The documentation describes the aggregation rule: "The overall verdict is binary, suspicious or valid, aggregated conservatively so that any suspicious detection makes the whole submission suspicious. The machine-learning services add confidence scores and visual explanations, while File Analysis contributes forensic findings from metadata, provenance, and file-structure examination" (docs.sensity.ai, read 9 November 2026). File Analysis is the layer the home page calls "forensic analysis", examining "every technical signature inside the file container, from codecs to timestamps" (sensity.ai, read 9 November 2026). Language does not apply, so the languages field is empty.

The vendor's history is on its about page, dated 3 February 2026: co-founder Giorgio Patrini was "a postdoctoral researcher working on deep generative models" at the University of Amsterdam in 2017, trained a deepfake detector in March 2018, and with Francesco Cavalli founded the company "late in 2018" (sensity.ai/about-sensity-ai, read 9 November 2026). The sibling video cards are Deepware Scanner, a free face-swap scanner, and Reality Defender; the one indexed media study scored Hive and AI or Not on images.

Limits and plans

Sensity plans with price, limits and the date each was checked
PlanPriceLimitsChecked on
Cloud platform and APITalk to sales; web app at platform.sensity.ai with teams and roles, REST API v2 with workspace-scoped keys, SDK, webhooks, CSV export; submission quota per billing period; no price published.
On-premiseServer setup on dedicated NVIDIA GPUs, workstation setup for a single investigator machine, or a USB dongle for field teams and air-gapped environments; no price published.

The documentation's limits page is the card's source for input limits, read 9 November 2026: files up to 500 MB by direct upload or public URL, audio and video up to 1 hour, video up to 2560 px on the longest dimension, images in jpg, jpeg, png, tiff, gif, webp and jfif, video in mkv, flv, mov, mp4, avi, webm, 3gpp and m4v, audio in wav, mp3, m4a, ogg, aac, flac and pcm (docs.sensity.ai/resources/formats-limits). URLs can be submitted from X, Instagram, TikTok, Facebook, LinkedIn, Reddit, Pinterest, YouTube, Vimeo, Dailymotion and Twitch. The submission endpoint "is rate-limited at the gateway to 100 requests per minute", and "each workspace has a submission quota for a billing period (monthly, quarterly, or yearly)"; on-premise deployments "do not enforce fixed media limits".

No price is published anywhere on sensity.ai or in its documentation, read 9 November 2026. The site's calls to action are "Get started", "Start a trial" and "Talk to an expert", and the documentation states the gate: "On Sensity cloud, a new signup is held for manual review. You cannot log in or create API keys until a Sensity administrator approves your account and you receive the welcome email" (docs.sensity.ai, read 9 November 2026). The card's free tier field therefore records none published, and the table below lists the two deployment forms the vendor describes without a figure in either.

Retention is stated once in years and otherwise open-ended. The privacy policy, last updated January 2024, reads: "We store the personal information we receive as described in this Privacy Notice for as long as you use our Services or as necessary to fulfill the purpose(s) for which it was collected", and then: "We will delete your biometric data within three years of your last interaction with the Services" (sensity.ai/privacy-policy, read 9 November 2026). The policy ties biometric data to "your selfie video and/or identification documents", and states no separate period for media files submitted for analysis; the documentation lets a workspace delete submissions through the API once processing ends. The card's retention rank field takes the three-year clause as the longest stated period, 1,095 days, and quotes the open-ended clause beside it. Ownership is Sensity B.V., Javastraat 25 I, 1094GZ Amsterdam, under terms last updated October 2026 and governed by Netherlands law with binding individual arbitration (sensity.ai/terms-and-conditions, read 9 November 2026).

What the vendor claims

No accuracy percentage is published. Home page: The Most Accurate Deepfake Detection Solution; title tag: Best Deepfake Detection Software in 2026; product page: Unparalleled accuracy in deepfake detection; By combining multiple detection techniques, we ensure higher accuracy, lower false positives, and robust adaptability against evolving threats. The 89.0% and 99.0% figures on the tech page are confidence scores in example result panels, not accuracy rates. Counters on the home page: 35 min saved on manual reviews per case and +30 countries where we are deployed.
sensity.ai, read

Sensity publishes no accuracy percentage. The home page headline reads "The Most Accurate Deepfake Detection Solution" and its title tag "Best Deepfake Detection Software in 2026"; the product page says "Unparalleled accuracy in deepfake detection" and that the multilayer approach "ensure[s] higher accuracy, lower false positives, and robust adaptability against evolving threats" (sensity.ai and product page, read 9 November 2026). None of those sentences carries a figure. The tech page shows "89.0%" and "99.0%" inside example result panels labelled Confidence, which are illustrations of a score, not accuracy rates, and the card records them as such (sensity.ai/tech-stack, read 9 November 2026). The home page's counters are "35 min saved on manual reviews per case" and "+30 countries where we are deployed".

The vendor's caveats sit in the documentation and the terms. The conservative aggregation rule means one suspicious service flags the whole file; "Final validation happens server-side. A submission that appears within limits can still be rejected on the server" (docs.sensity.ai, read 9 November 2026); and the terms provide the services "AS IS, WITHOUT WARRANTY OF ANY KIND" and do "not guarantee that it will be reliable or available 100% of the time" (sensity.ai/terms-and-conditions, read 9 November 2026). No false positive rate is published, and no statement is made about re-encoded or edited media.

What independent studies found

No record in the studies index tested Sensity, read 9 November 2026 in the data file. The study count is zero, and the false positive, false negative, robustness and claim gap fields are empty. The one indexed media record, Ha and Passananti 2024, scored Hive and Optic on 280 human artworks and 350 AI images and did not include Sensity (arXiv 2402.03214, read 9 November 2026); no indexed paper has scored any video or audio deepfake detector.

The vendor's forensic framing raises the stakes of that absence. The about page says the analysis "has to be conclusive and explainable enough to meet the test of evidence in a court of law" (sensity.ai/about-sensity-ai, read 9 November 2026), and the home page says the platform has been deployed "directly inside high-security forensic environments". A reader weighing that use has the vendor's account and, in this index, no outside measurement to set beside it.

Where Sensity AI appears in the index

Sensity ranks in one list, read 9 November 2026 in the data file: AI detectors ranked by data retention, on the stated 1,095 days from the three-year biometric clause, the second-longest period on that page after Grammarly Authorship's ten years. It is absent from the study-based rankings, which need an indexed record; from the free tier ranking, which finds no allowance here; from the language ranking, which does not apply; and from the claim versus evidence gap, because "Most Accurate" carries no percentage.

The card appears on AI detectors by content type, where video, image and voice are its columns and text is blank, on detectors with an API for the documented REST API v2, and in studies by modality with zero records against three non-text types. Every detector card on the register shows the same fourteen fields, so the empty price cells of an enterprise forensic platform can be read beside the published per-check rates of the consumer tools.

Sensity's On-premise plan places it among the fourteen cards on the offline question, added 9 November 2026, and its unnamed generator list places it among the six cards tabled apart on the generators covered ranking, added the same day.

Questions about Sensity AI deepfake detector

How much does Sensity AI cost?

No price is published: sensity.ai and docs.sensity.ai, read 9 November 2026, offer "Get started", "Start a trial" and "Talk to an expert" with no plan, quota or rate, and new cloud sign-ups are held for manual approval before login or API keys are issued.

What does Sensity AI detect?

Face manipulation (swaps, lip sync, reenactment, morphing), AI-generated images and video, synthetic and cloned voices, and file-level tampering, in files up to 500 MB and up to one hour long, per the product page and documentation read 9 November 2026. It does not analyse text.

How accurate does Sensity AI say it is?

It publishes no percentage: the home page says "The Most Accurate Deepfake Detection Solution" and the product page "Unparalleled accuracy", read 9 November 2026, and the 89.0% and 99.0% on the tech page are example confidence scores. No indexed study tested Sensity.

How long does Sensity AI keep data?

Personal information "for as long as you use our Services or as necessary", and biometric data deleted "within three years of your last interaction with the Services", per the privacy policy last updated January 2024 and read 9 November 2026; no separate period is stated for submitted media files.

Can Sensity AI be run on-premise?

Yes: the product page describes a server setup on dedicated NVIDIA GPUs, a single-workstation setup and a USB dongle for air-gapped field use, and the documentation says on-premise deployments do not enforce the cloud's fixed media limits, read 9 November 2026.

Sources and revisions

Sources

  1. sensity.aiRead 9 November 2026
  2. sensity.ai/deepfake-detection-for-video-i...Read 9 November 2026
  3. sensity.ai/tech-stackRead 9 November 2026
  4. sensity.ai/why-sensityRead 9 November 2026
  5. sensity.ai/about-sensity-aiRead 9 November 2026
  6. sensity.ai/privacy-policyRead 9 November 2026
  7. sensity.ai/terms-and-conditionsRead 9 November 2026
  8. docs.sensity.aiRead 9 November 2026
  9. docs.sensity.ai/resources/formats-limitsRead 9 November 2026
  10. arxiv.org/abs/2402.03214Read 9 November 2026

Revision log

  1. Card created2026-11-09: offline question and generators covered ranking linked (seventh wave)

Corrections log

Related reading