Sensity AI

• Every commercial figure Sensity publishes, from profitability and tripled revenue to the 98 percent accuracy measurement, is the company's own, so a buyer underwrites the trading record on Sensity's word.
• The 98 percent accuracy figure is measured on public data sets, so real-world performance against new generators could be materially lower and no independent test in the reviewed sources would show it.
• Sensity calls detection an arms race and says its foundational models close the gap against generators absent from the training data, so model work is a standing cost, and no reviewed source tests whether that approach holds.
• A biometric, fraud-prevention or identity-verification vendor already selling to these agencies and banks could add deepfake detection and push Sensity toward being one component of a larger purchase outside the forensic niche.
• On 3.2 million dollars of private capital, with the European grant and equity funding selected rather than committed, Sensity could be outspent on enterprise selling and model training if better-funded rivals consolidate the category.
• No inspectable certificate or audit report appears on Sensity's own probed pages, which can slow procurement at institutions that require one before deployment.

Sensity addresses a problem that is evidentiary before it is operational. Investigators, agencies and courts increasingly handle media that may be synthetic, and they need a result that survives legal challenge rather than a momentary flag. Europol put the same case in 2022 without any vendor involved, reporting that identifying deepfakes by hand does not scale and that law enforcement would need detection software built on AI, alongside examples running from chief-executive fraud to tampering with evidence.

The buyer is well defined and reachable. Sensity markets to intelligence units, law enforcement, digital forensics teams and prosecutor offices, and separately to banks, where it says its detection has been deployed on-premise inside identity verification pipelines and call-center systems to catch synthetic identities and voice cloning before a transaction is authorized. Sensity addresses the two groups with different material, court-ready evidence for the public-sector buyer and fraud screening for the bank, so the segments are distinct rather than one diffuse market.

What the record does not carry is an independent measure of the pain. Sensity's own fraud page documents the 2024 Arup case, in which deepfaked executives on a conference call led an employee to transfer more than 25 million dollars. That example is the company's own, and no source in the reviewed set sizes the market independently. [s14, s4, s6, s10]

The product is layered forensic detection across face manipulation, AI-generated images and video, and synthetic voices. Sensity describes an engine that reads pixel-level artifacts, acoustic patterns, metadata, behavioral cues and inconsistencies between one medium and another, stacking independent signals rather than resting the verdict on a single classifier, and its file examination traces generation history and modification traces to expose tampering.

Delivery is broad and aimed at agencies. Alongside a web application and an API, Sensity packages detection onto high-performance servers, offline workstations, and a portable USB dongle for verification in the field. Its documentation says the product runs cloud-hosted or on-premise, and an agency server deployment lets investigators share evidence across a private local network without an internet connection.

The output is what the company sells. The documentation says the machine-learning services add confidence scores and visual explanations while the file analysis contributes findings from metadata, provenance and file structure, so the customer receives evidence rather than a lone score, and the reports are generated in more than thirty languages with chain-of-custody controls. [s1, s3, s7, s4]

Sensity positions on forensic reporting built for legal use. Its own pages argue that a verdict is not enough and that analysis has to be conclusive and explainable enough to meet the test of evidence in court, and the company says this approach has let it repeatedly win public and private tenders against much louder companies that raised tens of millions in venture capital. That is the company's account of its own win rate, and no cited source corroborates it.

Rivals place Sensity in the same set. Resemble AI publishes a comparison page against Sensity and groups it with Pindrop and Reality Defender on its comparison navigation, framing its own product as securing the source through watermarking while casting Sensity as the tool that alerts you to deepfakes. No other comparative material appears in the reviewed sources, so a buyer weighing vendors is reading a competitor's own account of the difference.

The exposure is adjacency and verifiability. Biometric, identity-verification and fraud-prevention platforms sell to the same banks and governments and could fold detection into contracts they already hold, and because the reviewed sources name no Sensity customer, a buyer cannot yet confirm how deeply the product is embedded against that pressure. [s6, s18, s5]

Traction is stated in financial terms rather than named accounts. Sensity reports that a 2.1 million dollar round brought total capital raised to 3.2 million dollars and concluded a year in which it tripled revenue, reached profitability and achieved a capital efficiency ratio above 1.5x, with a goal of crossing 4 million dollars in annual recurring revenue during 2026. It also says it serves clients on four continents and is embedded in the forensic ecosystems of defense agencies, law enforcement, banks and insurers. Every one of those figures is the company's own.

Two records outside the company do exist. Biometric Update reported the round in its own voice, naming Auriga Cyber Ventures as lead with Betaworks Ventures and European angels participating, and the European Innovation Council's published list of selected companies names Sensity B.V. for blended finance, with the project described as a foundational security layer for deepfake detection. The list's own header notes that selection does not constitute a formal commitment for funding, so the record shows a decision rather than money received.

The motion is tender-led with a lighter product path underneath it. Sensity describes winning public and private tenders, which fits a procurement-driven sale, while a drag-and-drop web application and a documented API let an individual analyst or developer start small, though the documentation records that a new cloud signup is held for manual review until an administrator approves the account. [s8, s11, s13, s7, s1]

The founders pair research training with an early start. Giorgio Patrini co-authored the 2018 arXiv paper Sinkhorn AutoEncoders with Max Welling, and the arXiv record says the paper was accepted for oral presentation at UAI 2019. That is the one piece of the team's standing a reader can check away from Sensity's own pages, and Biometric Update names him as chief executive. The company's account adds that Patrini trained a deepfake detector in March 2018 and joined Francesco Cavalli, a threat intelligence specialist, before the company was founded late that year.

Longevity is the other signal. Biometric Update described Deeptrace in 2019 as an Amsterdam firm supplying deep learning and computer vision for detecting and monitoring synthetic media, naming Patrini as its founder and chief executive, so the record shows him on this problem years before deepfake fraud became a mainstream corporate concern.

What the cited record substantiates is research pedigree and a long operating history. It does not document a prior exit or a sustained publication run, and the case for commercial execution leans on self-reported profitability and tender wins rather than customers who speak publicly. [s5, s17, s11, s16]

Sensity's assurance story is built on deployment control rather than published attestations. On-premise servers, offline workstations and the USB dongle let an agency keep media inside its own boundary, and the company markets data sovereignty, private-network operation and chain-of-custody controls to buyers with strict data policies.

No inspectable certificate or audit report appears on the probed Sensity surfaces. A competitor's comparison page asserts that Sensity is GDPR and ISO 27001 certified without supplying either attestation, which is a rival's characterization rather than a document a buyer can read.

Court use is the assurance claim aimed at this buyer. Sensity markets reports it says are transparent, reproducible and admissible in judicial environments, and it sells courses and certifications that qualify analysts to produce and interpret the results. Those are product features built for legal scrutiny, and they do not answer the security questions a procurement team asks before deployment. [s19, s18, s4, s2]

Add analyzed competitors to compare them side by side with Sensity AI.

Sensity targets buyers who have to prove a piece of media is fake to somebody else, not merely flag it inside their own workflow. Government and judicial buyers are the segment the reviewed sources document most fully. Sensity markets to intelligence units, law enforcement, digital forensics teams and prosecutor offices, and Biometric Update describes the company as targeting law enforcement, forensics labs and governments.

Banking is the second documented segment. Sensity says its detection has been deployed on-premise inside banks, integrated into identity verification pipelines and call-center systems to catch synthetic identities, voice cloning and impersonation before a transaction is authorized. The company also lists insurers among the sectors it serves, but the reviewed sources document no insurance workflow, so the banking description is what the finance segment stands on.

The segmentation is disciplined around forensics. Sensity says it serves clients on four continents and is embedded in the forensic ecosystems of defense agencies and law enforcement, and the reviewed sources name none of them, so the breadth of the segment is asserted rather than shown.

The product detects face manipulation, AI-generated images and video, and synthetic voices through a layered engine. Sensity describes analysis across pixel-level artifacts, acoustic patterns, metadata, behavioral cues and inconsistencies between one medium and another, stacking independent signals rather than resting on a single classifier, and its file examination traces generation history and modification traces to expose tampering.

The claimed advantage is generalization plus explainability. Sensity argues that classifiers trained on specific generators lose accuracy against generators absent from their training data, and the funding Sensity was selected for is earmarked in part for foundational models meant to reduce that dependence. Its documentation says the machine-learning services add confidence scores and visual explanations while file analysis contributes findings from metadata, provenance and file structure, so the customer receives evidence rather than a lone score.

A cross-customer data advantage could sit inside the hosted web application if submissions accumulated into a shared detection corpus. The reviewed sources show no sign that Sensity retains customer submissions or has built such a loop, and offline and customer-run deployments limit what it would see.

The headline accuracy number is the company's own. Sensity states 98 percent accuracy tested on publicly available data sets, on its homepage and in its funding announcement, and no independent evaluation of that figure appears in the reviewed sources.

Sensity runs a tender-led sale into regulated and public-sector buyers. The company says its forensic approach has let it repeatedly win public and private tenders against much louder companies that raised tens of millions in venture capital, which fits a procurement-driven motion into agencies and financial institutions rather than a self-serve one. That win record is Sensity's own account and no cited source corroborates it.

A lighter product path sits alongside the enterprise sale. Analysts can drag files or URLs into a web application, and developers can integrate through a documented interface that pushes results to a webhook, though the documentation records that a new cloud signup is held for manual review until a Sensity administrator approves the account, so even the light path runs through the company.

The outside checks in this record are about funding. Biometric Update reported in its own voice that Auriga Cyber Ventures led a 2.1 million dollar round with Betaworks Ventures and European angels, and the European Innovation Council's published list names Sensity B.V. for grant and equity money on a project it describes as a foundational security layer for deepfake detection. The same list states that selection does not constitute a formal commitment for funding, and no reviewed source names a customer.

The reviewed pages publish no list price, which fits a negotiated sale to agencies and regulated institutions where deployment mode, scope and case volume drive the contract. The paths Sensity offers a visitor are a trial request and a conversation with sales, and a new cloud account is held for manual approval before any key is issued, so pricing sits inside the sales conversation.

The deployment menu shows where the pricing power sits. Cloud access suits lighter analysis, while high-performance servers, offline workstations and the USB dongle target agencies with strict data policies, and the cited pages put no price on any of them. Sensity also offers courses and certifications to the same accounts.

The reported capital efficiency hints at pricing discipline without establishing it. Sensity reports a capital efficiency ratio above 1.5x alongside profitability and a goal of crossing 4 million dollars in annual recurring revenue during 2026, and those are self-reported figures that say nothing about individual deal sizes.

Delivery spans a hosted web application, an interface for developers, and several offline modes. Analysts drag and drop files or URLs into the platform, developers embed detection through the API, and webhooks push results back as structured findings for other systems to consume.

The offline modes are the operationally demanding part. For agencies, Sensity packages detection onto high-performance servers for large-scale investigations, offline workstations for single investigators, and a portable USB dongle for verification in the field, and an agency server deployment lets investigators share evidence across a private local network without an internet connection.

Forensic output is the delivery contract. Reports are generated automatically in more than thirty languages with chain-of-custody controls, and Sensity offers courses and certifications that qualify a customer's analysts to produce and interpret the results. Delivery therefore ends with material an investigator can defend in a proceeding.

Sensity's assurance story is built on deployment control rather than published attestations. On-premise servers, offline workstations and the USB dongle let an agency keep media inside its own boundary, and the company markets data sovereignty, private-network operation and chain-of-custody controls to buyers with strict data policies.

No certificate or audit report a buyer can read appears on the probed Sensity surfaces. A rival's comparison page asserts that Sensity is GDPR and ISO 27001 certified without supplying either attestation, which is a competitor's characterization rather than a document a buyer can inspect.

Court use is the assurance claim aimed at this buyer. Sensity markets reports it describes as transparent, reproducible and admissible in judicial environments, and its certifications qualify analysts to use the platform in court proceedings and audits. Those are product features built for legal scrutiny, and they do not answer the security questions a procurement team asks before deployment.

Sensity positions as a detection and forensics layer that feeds other systems rather than a platform others build on. Its documented interface for developers lets banks and agencies route media through detection and pull structured results into their own fraud and investigative workflows, and the company describes that integration running inside identity verification pipelines and call-center systems.

Distribution beyond direct sales is thin in the public record. The web application, the developer interface and the offline packages reach different buyers, and the reviewed sources show no marketplace listing, no named technology partnership and no co-built product that would widen reach. The one outside relationship the record documents is with a funder: the European Innovation Council selected Sensity for grant and equity money, and it pairs its awards with business acceleration services that connect recipients to experts, investors and corporates.

The report format is the closest thing to an ecosystem asset. Sensity says investigators can export structured evidence packages for prosecutors, international partners and judicial authorities, so the output travels between organisations without a partner network behind it.

The founding team has research training and an early start. Giorgio Patrini co-authored the 2018 arXiv paper Sinkhorn AutoEncoders with Max Welling, and the arXiv record says the paper was accepted for oral presentation at UAI 2019. That is the part of the team's standing a reader can check away from Sensity's own pages, and Biometric Update names him as chief executive. Sensity's own history adds that Patrini trained a deepfake detector in March 2018 and joined Francesco Cavalli, a threat intelligence specialist, before the company was founded late that year.

Sensity was working on deepfake detection by 2018 and was covered publicly under its earlier name in 2019. Biometric Update described Deeptrace that year as an Amsterdam firm supplying deep learning and computer vision for detecting and monitoring synthetic media, naming Patrini as its founder and chief executive, so the record shows him on this problem for years.

The cited record substantiates research pedigree and a long operating history. It does not document a prior exit or a sustained publication run, and the case for commercial execution leans on self-reported profitability and tender wins rather than customers who speak publicly.