How do you detect a deepfake video of an Indian CEO?

As of 24 October 2026: the service descriptions, prices and turnaround figures on this page are current.

What forensic signals expose a deepfake video?

Six, and they are strongest in combination. Blink cadence: generated faces either do not blink or blink in a rhythm that does not match speech. Lip sync drift, most visible on plosives and on words with closed lips, where the mouth shape lags or overshoots the audio. Skin texture: a generated or enhanced face loses the micro-texture and the specular pattern of real skin, and the forehead and cheek areas are where this shows first. Jaw and neck boundary: composite artefacts concentrate where the face meets the neck, the hairline and the collar. Background warping: generative fill distorts edges and reflections behind the head, especially in a modern office with glass. And compression behaviour: a real recording survives re-encoding more gracefully than a generated one, because the generative artefacts are high-frequency and re-encode badly. Detection tools score all of these, and no single score is a verdict, which is why we treat the numeric result as one input among several.

Why did the I4C warn banks about deepfakes in June 2026?

Because the control banks relied on stopped being a control. The I4C advised that deepfake video and synthetic identities are being used to defeat facial authentication, liveness detection and video-based KYC, with contact typically beginning on social media, messaging apps, job portals, dating sites and phone calls. The technique is not exotic: recorded blinking, head turns and focus changes are captured and replayed, and pre-recorded footage is fed into a live video call. The honest conclusion for a risk owner is that video KYC is one factor, not two, and that a control whose only job is to prove a face now proves much less than it did two years ago. Any Indian bank or fintech that relies on a liveness check as a second factor needs a compensating control, and the compensation has to be procedural.

What is the SEBI caution of 17 July 2026 about?

It is about the pattern, not the technology. SEBI cautioned listed companies that fraudsters impersonate CEOs and managing directors over email, WhatsApp and video calls to push finance staff into urgent transfers, and it described two methods: external impersonation using AI voice cloning and deepfake video, and internal compromise where malware hijacks the victim's active WhatsApp Web session. What SEBI asked for was procedure, not detection technology: verification steps independent of the communication channel, and a rule that no single person can release a new beneficiary or trigger a large transfer on their own. The reason the video itself is not the answer is that a convincing deepfake defeats the eye, and the control has to live in a workflow that the eye does not sit inside.

What should an Indian finance team do on receiving a video call from the CEO?

Four steps, in order. First, do not act on the call. Whatever is requested, the request needs to leave the channel it arrived on. Second, verify out of band: call a number already on file for that executive, not one provided in the message, and not one dialled from the WhatsApp profile. Third, apply dual authorisation: no new beneficiary and no transfer above your internal threshold moves on one approval, and the second approver is someone outside the requesting chain. Fourth, apply a cooling-off period on any new beneficiary or changed bank detail, ideally 24 to 48 hours, which alone stops a large share of these cases because the fraud window closes. Record the call if policy allows, keep the message thread, and escalate to CERT-In within 6 hours if an incident occurs.

What are the frequently asked questions?