Sensity Commits EIC Accelerator Money to Deepfake Detection

Before Sensity sold deepfake detection, it counted deepfakes. In October 2019, under its first name Deeptrace, the Amsterdam company published The State of Deepfakes, which Vice reported found 14,678 deepfake videos online, 96% of them non-consensual pornography; the company still hosts the summary on its Medium page. Seven years later the European Innovation Council has selected it to build what the project calls a "foundational security layer for deepfake detection".

CORDIS records an EU contribution of €2,479,190 to Sensity B.V. for a 24-month project from 1 October 2026 to 30 September 2028, under the call HORIZON-EIC-2026-ACCELERATOR-02, with the grant agreement signed on 23 September 2026. The project is five weeks old. This article therefore sets out what was promised and what can already be checked, not what was delivered.

What the EIC committed

The EIC announced on 15 June 2026 that it had selected 38 companies from 87 interviewed, with €90 million in grants and €202 million provisioned for equity; its announcement names no company in the text. The Dutch enterprise agency RVO listed Sensity among three Dutch start-ups sharing €22.6 million, and wrote that with EIC support the company is scaling its technology to image, audio and video.

Biometric Update reported on 17 July 2026 a package of €4.9 million, of which €2.5 million is grant. CORDIS records only the grant, at €2,479,190, and lists no total project cost. No equity investment by the EIC Fund in Sensity has been published, so the equity part is a commitment in the selection, not a recorded investment. The same report gives a claim from the company that it was the only cybersecurity solution selected, and a figure of 1,760 proposals that the EIC announcement does not contain.

The pitch: detectors that do not go stale

Most deepfake detectors are trained on output from particular generators and, by Sensity's own account to Biometric Update, lose accuracy on media from generators they have not seen, which makes them brittle and costly to retrain. The company says its approach "decouples the learning of forensic features from the classification of specific generators". The EIC money, per the same report, is for those foundational models and for real-time analysis on constrained hardware.

The CORDIS objective describes a platform that combines visual and audio deep learning, forensic analysis, explainable AI and cyber threat intelligence, deployed in the cloud or on air-gapped machines on site. Sensity's government page offers server, workstation and USB dongle versions, reports in more than 30 languages, and "forensic-grade evidence suitable for court". It names no agency that uses it, and no court ruling on such evidence could be found.

The project's own case rests on the figures above: deepfakes in 0.1% of financial fraud attempts in 2021 and 6.5% in 2024, a rise the summary puts at 2,137%. The objective does not cite the study behind them.

What can already be checked

The most widely reported part of Sensity's record is its research on deepfake abuse. In October 2020 it reported that a Telegram bot had been used to create fake nude images of about 104,852 women, a figure reported by The Star in Kenya; The Star in Malaysia put it at more than 100,000. In May 2022 it published an attack test on ten unnamed identity verification vendors, reported by Biometric Update and the Daily Dot. Patrini said the vendors had been told months earlier and had downplayed the results; none of them is quoted on the record.

In April 2026 Euronews reported that Sensity had found more than 1,000 similar videos of Ukrainian soldiers on social media, with a core set of 60 manipulated by AI, first posted on TikTok or Telegram from new or anonymous accounts; inkl carried the same findings.

The one outside test found is a New York Times comparison of more than a dozen detectors across more than 1,000 scans, republished by The Star in March 2026. It found that Sensity and Resemble.ai did best among seven tools checking audio. The republished text does not name Sensity in its image results. Sensity's product page promises "unparalleled accuracy in deepfake detection" but gives no figure or method, and no independent accuracy audit of its detector was found.

1. Foundational detection models that generalise to unseen generatorsToo earlyProject started 1 October 2026; CORDIS shows no report or result yet.
2. Real-time analysis on constrained hardwareToo earlyStated aim in Biometric Update's report; no benchmark published.
3. Detection across image, audio and videoProduct existsAll three modes offered on the company site; audio performed well in the New York Times test.
4. Evidence fit for courtClaimCompany statement; no court ruling or named agency found.

The company behind it

A summary of Sensity's filed accounts on liza.nl, whose VAT number matches the one CORDIS gives for the beneficiary, shows a balance sheet total of €2,292,726 for 2025 against €475,035 for 2024, with equity of €1,200,197 and debts of €1,092,529 in 2025. The summary shows no turnover or result, and the filed accounts themselves were not obtained for this article. Revenue and profitability figures circulating on startup database sites conflict with one another and have no primary source, so they are left out here.

The $2.1 million round led by Auriga Cyber Ventures is reported by Biometric Update as closed earlier in 2026 and listed by Nordic9 without a date; no announcement from Sensity or Auriga could be found. Patrini has led the company as co-founder since its Deeptrace days, as a 2020 interview shows, and Euronews quotes Cavalli as Sensity's founder.

Why Europe funds it

The EU's AI Act requires anyone deploying an AI system that generates or manipulates image, audio or video content constituting a deepfake to disclose it, under Article 50 of the regulation. Disclosure depends on deployers complying; detection addresses content whose makers do not, and Sensity told Biometric Update it targets police, forensic laboratories and governments. Patrini told Biometric Update that the EU support reaffirmed "tech sovereignty" in a critical security sector. The Ukraine analysis shows the kind of material such customers are meant to identify: synthetic video produced at volume and, per Sensity, aimed at morale.

The reporting is based on publicly available sources, listed below.