Deepfake and impersonation
A file check and guidance only. No detection engine runs, so Sentinel never says whether media is real or fake.
- Available
- 01
- Limited
- 02
- Not released
- 01
- Coming
- 02
How the attack works
Voice cloning and face-swap tools need only a short sample of someone's voice or a few photos. The attack is still social engineering — "it's me, I'm in trouble, send money now" — but the familiar voice or face removes the doubt that would normally make you check.
- STAGE 01
Sample
Audio or video of the target is taken from social media, voice notes or public videos.
- STAGE 02
Synthesis
A cloned voice or a face-swapped video is produced, often good enough over a phone line or a low-quality video call.
- STAGE 03
Emergency
A call from a "relative" in trouble, or a "manager" approving an urgent transfer.
- STAGE 04
Payment
Money is sent before anyone calls back the real person.
- Parents and grandparents
- Finance staff who act on a senior manager's instruction
- Public figures whose voice is widely available
A common pattern: a parent receives a call from an unknown number in what sounds like their child's voice, crying and asking for money for an emergency, then a second person takes the phone and gives payment instructions and insists on secrecy.
Cloned voices and edited video make "it's me, send money now" calls convincing. No tool can tell real from fake with certainty.
Sensor, rule, evidence, action
Each step carries its own state. A step that is not built is shown as coming, not skipped.
File check
Pick an image, audio clip or video. Sentinel reads what the file says about itself (type, size, metadata, Content Credentials shown but not verified). The file is not kept. There is no real-or-fake analysis.
LimitedAndroid 4.50, web portalVoice-clone and impersonation signals
Signals that a voice or identity is being imitated.
ComingProbability with method
A result will be a probability with the method and its confidence — never "100% real" or "100% fake".
ComingProvenance only when proven
Signed-media provenance will be shown only where it is cryptographically verified.
Coming
When it ships, a result will be a probability with the method and its confidence — never "100% real" or "100% fake".
Provenance (signed media) will be shown only where it is cryptographically proven.
Until then, the practical defence is a family code word and calling the person back on a number you already have.
What works today, and what is coming
What works today
03In production or in the current download.
- File check for pictures, audio and videoLimited
Shows what the file says about itself (type, size, metadata, whether it carries Content Credentials, which are shown, not verified). It never says "real" or "fake". The file is not kept.
Android app 4.50, web portal
- Impersonation and voice-clone guidanceLimited
What to ask, a family code word, and calling the person back on a number you already have. Guidance, not detection.
Android app 4.50
- File check historyAvailable
Past checks, kept on the phone.
Android app 4.50
Built, not yet released
01The code exists and is tested in the repository; it has not shipped to users.
- Deepfake provider interfaceNot released
A provider contract and tests, plus callback and family code-word help. No configured analysis provider means no automated deepfake verdict. Sentinel does not listen to live calls.
Coming
02Not built yet. Nothing on this list runs today.
- Real-or-fake analysis of mediaComing
A detection engine for edited or generated media.
Premium
- Voice-clone and impersonation detectionComing
Signals that a voice or identity is being imitated.
Premium
2 attack families map to this capability
Copied row by row from our engineering register. A family counts as detected only when a rule, a sensor that feeds it and a test that makes it fire all exist.
Feature states above and register statuses measure different things: a feature can be built while the family it addresses stays unobserved, because the register counts sensors and rules, not screens.
| # | Attack family | Status | What the register says |
|---|---|---|---|
| 036 | Vishing | No sensor | No call telemetryNothing runs |
| 039 | Deepfake Impersonation | No sensor | No media analysisNothing runs |
What runs where, and what never leaves
Classification happens on the device before anything is queued. Cloud reporting is off by default on consumer devices and turning it off also stops events already queued.
Runs on the device
- Nothing today: no media analysis runs
May reach the cloud
- Nothing today
Never leaves
- Your media is not collected today; any future upload will be an explicit check you start
Editions that include it
Each feature above names the platform or edition it runs on. An edition's own page lists the full state of that edition.
What to do if…
Practical steps that work whether or not you use our products.
A family member calls in distress asking for money
- 1.Hang up and call them back on the number you already have.
- 2.Ask a question only the real person would know, or use a family code word agreed in advance.
- 3.Do not let anyone keep you on the line while you pay.
A "manager" asks for an urgent transfer by call or video
- 1.Verify through a second channel you already trust.
- 2.Follow the normal approval process; urgency is a red flag, not a reason to skip it.
Your face or voice is used in a fake
- 1.Report the content to the platform hosting it.
- 2.Tell family and colleagues so they know to verify calls.
- 3.In India, report on cybercrime.gov.in.
Limits and known gaps
- GAP-01File checks, verification help and a tested provider interface exist. Without a configured analysis provider there is no automated deepfake or voice-clone verdict; Sentinel never listens to live calls.
- GAP-02The register records deepfake impersonation and vishing as not observable.
- GAP-03When built, no detector will be able to tell real from fake with certainty; results will be probabilities.
Frequently asked
Can I check a video today?
What should I do in the meantime?
Capabilities that work alongside it
Checks a message, payment request or caller against known scam patterns, on the device.
Password exposure checks, account sign-in safety and hacked-account help today; verified-address monitoring interfaces need a configured provider for results.
Defence against prompt injection, agent hijacking and poisoned models. Not built yet.