Checks photos, recorded clips and live video for face-swaps, re-enacted faces and fully generated people before they are trusted by a KYC officer, an editor or an investigator.
How it works, in general
- Frames are sampled and the face region is isolated.
- Models look for blend seams, texture and lighting that do not fit the scene, and lips that do not match speech.
- Signals are combined into a verdict; flagged media opens a case with the evidence kept on the box.
Use cases
- Video KYC and V-CIP sessions
- Viral clips before a newsroom runs them
- Digital evidence triage for cyber cells
- Remote job interviews and board calls
Confirm with FaceOff* supported formats and resolutions · live stream vs file · results on your own sample set · streams per box
Checks that the person using a session behaves like the real account holder, so a stolen password, OTP or remote-access session is not enough to take over an account.
How it works, in general
- The system learns a person's usual rhythm: typing, swiping, pointer movement and navigation.
- Each new session is compared continuously, not just at login.
- A mismatch triggers step-up verification or blocks a risky action.
Use cases
- Net-banking and app account takeover
- Mule accounts run by one operator
- Admin and treasury consoles
- Call-centre agent desktops
Confirm with FaceOff* web, Android and iOS collection · learning period · how behavioural data is minimised. Behaviour that identifies a person is personal data under the DPDP Act, so plan notice and purpose up front.
Flags calls and voice notes made with cloned or generated voices, such as a "CFO" approving a payment or a "relative" asking for urgent money.
How it works, in general
- Audio is analysed for traces that synthesis tends to leave: odd spectral patterns, too-steady pitch, missing breaths, channel seams.
- Where a person has consented, the call can be compared with their enrolled voice.
- The result routes the call: carry on, call back on a known number, or stop.
Use cases
- Contact-centre caller checks
- CEO and vendor payment fraud
- Helplines and grievance desks
- Voice evidence for investigations
Confirm with FaceOff* Indian languages and accents · telephony and SIP integration · minimum audio length · real-time vs recorded
Treats every access request as untrusted until checked, instead of trusting a session for hours after one login.
How it works, in general
- Each request is checked against who the user is, the device, behaviour and context such as time and location.
- Low-risk requests pass quietly; risky ones ask for more proof.
- Every decision is logged for audit and incident response.
Use cases
- Privileged and admin users
- Remote staff and vendors
- Control rooms and sensitive registries
- Approvals for high-value transactions
Confirm with FaceOff* identity provider and SSO connectors · MFA methods · policy model. See also XcellHost AI Identity Security.
Watches public sources for fake profiles of your leaders, lookalike domains, fake customer-care numbers, leaked data and deepfakes that use your name.
How it works, in general
- Public sources are collected: social media, websites, domain registrations, app stores and open forums.
- Findings are matched to your brands, people and assets, and ranked by risk.
- Each finding becomes a case with evidence and a suggested next step.
Use cases
- Executive and brand impersonation
- Fake helpline numbers and apps
- Police and agency intelligence
- Early warning for the SOC
Note. Collecting open sources needs controlled outbound access; analysis stays on the box*. Confirm with FaceOff* source coverage · languages · takedown support.
Turns camera feeds into alerts and searchable events, processed on the appliance rather than sent to an outside cloud.
How it works, in general
- Camera streams are read on your network.
- Models detect people, objects and events you define, such as entry to a restricted zone.
- Alerts go to your control room; footage and metadata stay on site.
Use cases
- Campuses and data centres
- Bank branches and ATMs
- Public spaces and events
- Critical infrastructure sites
Confirm with FaceOff* streams per box · camera protocols · retention settings. CCTV images of people are personal data; signage and retention rules apply.
Gives analysts and officers a plain-language assistant without sending case data, faces or call recordings to a public AI service.
How it works, in general
- An analyst asks a question: why was this flagged, what links these cases, draft the incident note.
- The small language model answers on the appliance, using case data that is already there.
- Answers and prompts are logged on site for review.
Use cases
- Air-gapped and sovereign SOCs
- Police and government case desks
- BFSI fraud and risk teams
- First drafts of CERT-In reports
Confirm with FaceOff* languages · model updates without internet · hardware sizing.