Watch Us Catch a Deepfake Live at VivaTech 2026
UncovAI is at VivaTech 2026 — Paris, June 17–20 — running live on-device deepfake detection at the BNP Paribas and Caisse des Dépôts booths. No slides. No recorded clips. The real thing, running in front of you.
The threat is already in your meetings
Deepfakes aren't a threat companies plan for in the future. They're happening now — in live video calls, in WhatsApp messages impersonating executives, in synthetic faces authorizing transactions that nobody questioned because they looked real.
Most organizations still treat deepfake risk as a perimeter problem, something that happens at the edge of their network. The actual attack surface is narrower and closer than that: it's the video call your CFO just joined, the voice note from a supplier, the document your legal team signed off on.
Cloud-based deepfake detection sends your sensitive communications to a server you don't control. The interception risk introduced by the detection pipeline can exceed the risk of the deepfake itself.
UncovAI's real-time deepfake detection was built around one constraint: your data never leaves your infrastructure. Detection runs on the device where the communication is happening. No cloud round-trip. No third-party model processing your executive calls. Just the result, computed locally, delivered in the same moment the video arrives.
What we're showing at VivaTech
Everything running at our booth is live — the same technology we deploy for enterprise customers, operating on real input in real time.
Live deepfake detection
Frame-by-frame synthetic face analysis on live video. On-device — no latency compromise, no cloud round-trip.
Lip-sync & gaze tracking
Proprietary models that catch the micro-desyncs between voice and lip movement, and the gaze deviations AI-generated faces can't hide.
AI-generated image & text detection
Synthetic photograph classification and AI-written document analysis — the full surface area of generated content.
Secure messaging deployment
Detection running inside WhatsApp, Telegram, and enterprise video conferencing. Zero third-party dependencies, entirely within your infrastructure.
Our video detection stack and image detection models will both be running during the demos, so you can see how each layer behaves on different types of synthetic content.
Why "sovereign AI security" is not marketing language
Every cloud-based detection service makes an implicit offer: give us your video and we'll tell you whether it's real. For consumer use cases, that's a reasonable trade. For organizations handling regulated data, M&A communications, or government decision-making, it isn't.
A detection pipeline that routes your communications through a third-party server is a new attack surface, not a mitigation. The data governance risk is structural, not a configuration problem you can policy away.
No cloud. No black box. Sovereign AI security built entirely on your terms.
Our architecture runs the full detection stack — face authenticity, lip-sync, gaze tracking, image classification, text analysis — on the hardware where the communication is happening. The model never sees the outside world. This is what makes enterprise-grade deepfake and AI fraud protection viable for organizations that can't accept third-party data exposure as a term of service.
What the demo looks like
The demo runs in about ten minutes. We'll walk you through a live video stream being analyzed on the device in front of you — showing the lip-sync model flagging desyncs invisible to the naked eye, the gaze tracking system catching the deviation patterns that define AI-generated faces, and an AI-generated image being classified in under a second with the features that triggered detection shown clearly.
After that: the questions that matter to your organization. How does it deploy in your existing infrastructure? What does on-premise installation require? How does it integrate with the tools your teams already use?
If you're evaluating whether this fits your use case, the demo is a faster path to an answer than any documentation.
Common questions
What exactly does UncovAI's deepfake detection do?
UncovAI analyzes live video streams on-device in real time using proprietary computer vision models. It detects synthetic faces, lip-sync inconsistencies, and gaze anomalies without routing any data to a cloud server. Everything runs locally — on the device or on-premise server handling the communication.
How is on-device detection different from cloud-based services?
Cloud-based services require your video data to leave your infrastructure and be processed by a third-party server — introducing data sovereignty risk and a potential interception vector. UncovAI runs the entire detection model within your infrastructure. No data is transmitted externally at any point.
Can it detect deepfakes on WhatsApp and Telegram?
Yes. UncovAI deploys inside WhatsApp, Telegram, and enterprise video conferencing applications. Detection runs on-device with zero third-party dependencies — no audio or video data is ever routed to an external server during analysis.
What types of AI-generated content does UncovAI detect?
The full stack covers real-time deepfake video, AI-generated images, AI-written text and documents, synthetic voice, and lip-sync manipulation. Each detection layer runs independently and can be deployed modularly based on an organization's specific threat surface.
Who is this for?
Primarily organizations where the cost of a successful identity attack is high: financial institutions, legal firms, government bodies, healthcare providers, and any enterprise where executive communications or authorization chains are targets. The technology scales from individual workstations to enterprise-wide on-premise deployments.
Come find us at VivaTech 2026
BNP Paribas booth, June 17–20. Caisse des Dépôts, Pavillon 7, Stand 2D48, on June 17. Ten minutes to see it run. The conversations after tend to go longer.
Follow us on LinkedIn →