Enterprise security
Warn staff when a video in the browser looks synthetic, reducing social-engineering risk.
How the sector deploys itDeepBrowserGuard
A browser extension that quietly checks the videos you watch online for deepfakes and synthetic media. It works across websites in the background, flagging manipulated or AI-generated video as it plays — without slowing your browsing down
Product Walkthrough
0:000:00
What it is
It checks the video you are already watching
Most ways of testing a video ask you to leave the page: download the clip, find a checker, upload it, wait. DeepBrowserGuard removes that trip entirely. It is a browser extension, so the check happens where the video plays, while it plays.
A content script watches each page for video as it appears — HTML5 players and WebRTC streams, including cross-origin iframes and shadow DOM. Frames are sampled on the device, analysed off it, and the verdict comes back on screen in tens of milliseconds.
Your screen and your video never leave the device. Only small, localised bounding-box data is sent for analysis.
- Average round-trip to the analysis nodes and back
42 ms
Average round-trip to the analysis nodes and back
- Adaptive frame sampling on the device, WebP compressed
1–5 FPS
Adaptive frame sampling on the device, WebP compressed
- In-page memory footprint, so pages stay responsive
Under 2 MB
In-page memory footprint, so pages stay responsive
- Only localised bounding boxes leave the device
GDPR · SOC 2
Only localised bounding boxes leave the device
Try it here
Let’s evaluate AI-generated Deepfake Video here
Drop a video or image here
MP4, WebM or MOV up to 100 MB · WebP, PNG or JPG
The check starts as soon as a file lands · 1 credit
Sample Media
1 credit per check
Run the same check the extension runs — on your own file, on a page you were looking at, or on one of ours.
12 / 12Free checks left
See credit termsHow it works
From a video on the page to an on-screen verdict in milliseconds
01
Browser video
HTML5 and WebRTC streams detected on the page
02
Content script
Samples frames and sends only bounding boxes
03
Secure stream
Encrypted persistent link, sub-50 ms round-trip
04
Analysis nodes
GPU inference flags synthetic media
05
On-page verdict
Result shown live; activity logged for admins
What sets it apart
Ordered by impact, most decisive first
Real-time detection without the lag
It analyses video streams as they play, using asynchronous processing and adaptive frame sampling so there is no buffering or stutter. Protection runs in the background while you keep watching normally — you get a verdict, not a wait.
Flagship capability · no buffering, no stutter
Privacy-first by design
Rather than sending your screen or the full video, it transmits only small, localised bounding-box data for analysis. This zero-trust approach to privacy is built to line up with GDPR and SOC 2 expectations.
Data minimisation · bounding boxes only
Low-latency streaming backend
A persistent encrypted connection to the analysis nodes keeps round-trips under 50 ms, so verdicts arrive fast enough to matter while a clip is still on screen.
About a 42 ms average round-trip
Works across the whole web
Intelligent content-script injection watches each page for new video as it appears — HTML5 players and WebRTC streams, including cross-origin iframes and shadow DOM — so media is caught wherever it shows up.
HTML5 · WebRTC · iframes · shadow DOM
Enterprise deployment via MDM
Roll it out across an organisation through device-management policies, with admin reporting and local site whitelisting so trusted internal pages can be excluded.
Policy rollout, central reporting, whitelisting
Secure, efficient telemetry
Frames travel over a TLS 1.3-encrypted pipeline using a compact binary format, with automatic failover and region-based routing to keep detection reliable and responsive.
TLS 1.3 · compact binary · region routing
How it compares
Measured against what people actually do instead
Checks video where you watch it
- Upload-and-scan tools
- No — Download then upload
- Spotting it yourself
- Partly — On the page, unaided
- DeepBrowserGuard
- Yes — In the browser
Works in real time, as it plays
- Upload-and-scan tools
- No — After the fact
- Spotting it yourself
- Partly — Your judgement
- DeepBrowserGuard
- Yes — Yes
Automatic on every video
- Upload-and-scan tools
- No — One upload at a time
- Spotting it yourself
- No — No
- DeepBrowserGuard
- Yes — No manual steps
Covers embedded and cross-origin video
- Upload-and-scan tools
- Partly — Depends on file
- Spotting it yourself
- No — No
- DeepBrowserGuard
- Yes — HTML5 + WebRTC, iframes
Privacy-preserving
- Upload-and-scan tools
- Partly — Full file uploaded
- Spotting it yourself
- Yes — Nothing sent
- DeepBrowserGuard
- Yes — Sends only bounding boxes
Enterprise rollout via MDM
- Upload-and-scan tools
- No — No
- Spotting it yourself
- No — No
- DeepBrowserGuard
- Yes — Policy-based
Central admin reporting
- Upload-and-scan tools
- Partly — Varies
- Spotting it yourself
- No — No
- DeepBrowserGuard
- Yes — Yes
The architecture
Found on the page — judged off it
One boundary, and only bounding boxes cross it
- HTML5 players
- WebRTC streams
- Iframes · shadow DOM
On the device
Content script
Samples frames and reduces them locally
Trust boundary
Only bounding boxes cross. Your screen and your full video stay on the device.
Off the device
Analysis nodes
Flags manipulated and AI-generated video
≈42 ms
On-page verdict
Use cases by sector
Where an in-browser check changes the outcome
Banking & finance
Flag fake executive or advisor videos used in investment and payment scams.
How the sector deploys itMedia & journalism
Give reporters a first-pass check on clips sourced from the open web.
Government & public sector
Help officials spot manipulated video during normal browsing.
How the sector deploys itDeployment strategy
Install once, protected everywhere you browse
Install the extension
Add DeepBrowserGuard to a Chromium-based browser such as Chrome or Edge. For organisations, push it to whole fleets through MDM policies.
It watches the page
A content script uses the browser's mutation observer to hook into video elements and WebRTC streams as they appear — including cross-origin iframes and shadow DOM.
Adaptive sampling on-device
Frames are sampled at 1–5 FPS with light WebP compression, keeping the in-page memory footprint under about 2 MB so pages stay responsive.
Secure streaming to analysis nodes
Frames stream over a TLS 1.3 encrypted channel to accelerated inference nodes, with region-based routing and automatic failover for reliability.
Instant on-page verdict
Results return in tens of milliseconds — around a 42 ms average round-trip — and surface right where the video plays, while admins can review activity centrally.
Four things a browser check has to get right
Where it runs, what it does with your data, what it costs the page, and what an administrator can do with it.
- Where it runs
- A Chromium-based browser extension for Chrome and Edge. The client-side footprint stays under roughly 2 MB of DOM overhead, so everyday browsing feels unchanged.
- Privacy model
- Strict data minimisation — only localised bounding-box data leaves the device, aligned with GDPR and SOC 2. Your full video and screen stay with you.
- Performance
- Adaptive 1–5 FPS sampling with WebP compression on the client, and sub-50 ms streaming latency — about a 42 ms average round-trip — to the analysis backend.
- Enterprise controls
- MDM-based deployment across managed devices, central admin reporting, and local site whitelisting so trusted internal pages can be left out of analysis.
Frequently Asked Questions
The extension analyses streams as they play using asynchronous processing and adaptive frame sampling — no buffering or stutter. Frames are sampled at 1–5 FPS and compressed on-device, then streamed over a persistent, secure real-time connection to the analysis nodes, keeping round-trips under about 50 ms (roughly 42 ms average). Results surface on-page where the video plays, within tens of milliseconds.
No. It is privacy-first by design: only small, localized bounding-box data leaves the device — not the full video — and the pipeline is TLS 1.3-encrypted. The DOM footprint stays under roughly 2 MB. This model is aligned with GDPR and SOC 2.
Chromium-based browsers (Chrome, Edge). It works across the whole web, detecting standard and live-streaming video players, including those in cross-origin iframes and content embedded deep in the page.
Yes. It can be rolled out through device-management (MDM) policies, with admin reporting and local site whitelisting. Detection activity is logged centrally for admin review. The telemetry pipeline uses a compact, efficient format with automatic failover and region-based routing.
Enterprise security teams (reducing social-engineering risk), banking and finance (flagging fake executive or advisor videos in scams), media and journalism (a first-pass check on web-sourced clips), government, education (media-literacy programs) and online-platform trust-and-safety teams.
The rest of the line
More in SyntheticMediaGuard
Check a video in the browser you already use
Run one on this page, or have a solutions engineer walk your team through a rollout across managed devices.