Every format an attacker can use
Voice, image, video and documents analysed in one platform, with one history and one set of reports, because a real incident never arrives in a single format.
Six products,
one platform
Each one has its own page with what it catches, how it runs and what an evaluator needs to check.
From media to verdict in milliseconds
Voice clones now take seconds of reference audio, and generated images and video arrive in volume. The same three steps answer all four formats, so an incident is not four separate investigations.
Step one
Submit or stream
Upload audio, images, video or documents, paste a link, send it over WhatsApp, or let Raid AI join a live call.
Any format, any channelStep two
Analyse
Frequency and timing for audio, pixel artefacts for images, frame coherence for video, forgery checks for documents.
Checked the way each format is fakedStep three
Verdict and report
A confidence scored judgement while the call is still open, then a timestamped report and an alert to your SOC.
While the call is still open
Detection where
the conversation already is
Nobody stops mid incident to open a portal. Verdicts arrive in the tools people are already holding.
X bot
Mention the bot on a post and it analyses the media in the thread.
- Public detection on X
- Audio, image and video in one reply
- Multiple detectors behind each verdict
Link scanning
Paste a link and the media behind it gets pulled and analysed.
- Posts on X and videos on YouTube
- Any page carrying a preview image
- Direct media files
The questions
security asks second
Accuracy gets you into the evaluation. These are what decide it.
Zerodata retained after analysis
Retention you control
Choose zero retention, where media is discarded once the verdict is returned, or a window you set. None of it trains a model.
Append onlyaccess, exports and configuration
Every action is logged
An append only audit trail covering access, exports and configuration changes, streamable to your SIEM.
120+languages, Arabic dialects included
Multilingual by design
Detection in over 120 languages, and the best Arabic deepfake detection model in the world, first on ArA-DF 2026.
Per userlimits drawn from the organisation's allowance
Roles and spending caps
Granular permissions, plus a limit for each person, drawn from the organisation's allowance, so one account cannot use it all.
Runs in your infrastructure
Run the full detection stack on premises, inside your own network, so media and verdicts never leave your perimeter.
Isolated per tenant
Enterprise tenants get a dedicated database rather than a shared table with a filter on it.
Your data stays yours
Export everything your tenant holds through the API, on demand, without raising a support ticket.
Shareable reports
Send a findings link to someone outside the platform, or print it, with the access recorded.