Spot the three dominant deepfake scam patterns in 2026 — celebrity endorsement videos, CEO video calls, and romance / partner deepfakes — with the visual and audio signals that defeat them.
Last reviewed: 13 May 2026 · ScamSupport research
Why deepfake scams crossed the consumer threshold in 2026
Two technical thresholds were crossed in 2024-2025 that changed the deepfake landscape. First, generative video systems became good enough to fool casual viewers in short clips (under 60 seconds). Second, the cost of producing them collapsed — a working deepfake of a public figure can now be assembled in hours from open-source toolchains, at near-zero marginal cost.
The result is that deepfake video is no longer the rare hand-crafted artefact of 2022. It is commodity output of organised scam operations. UK consumer victims of deepfake-driven fraud are surging in Report Fraud reporting, particularly in three categories: deepfake-endorsement crypto-platform ads (Martin Lewis, Elon Musk, Richard Branson are the dominant impersonated figures), deepfake business-email-compromise (BEC) over video calls, and deepfake romance / pig-butchering with intermittent “video proof” calls.
The good news: most current deepfakes are detectable by a careful human viewer if you know what to look for. The systems are improving, but the detection signals below remain reliable for the bulk of 2026 deepfake content.
Three deepfake scam variants currently in circulation
Variant 1 — Celebrity endorsement video (Martin Lewis, Elon Musk, Richard Branson, This Morning hosts)
How it presents: A short video (30-90 seconds) appears as a sponsored ad on Facebook, Instagram, YouTube, TikTok or via a forwarded link in WhatsApp / Telegram. The video features a recognisable UK or international financial figure — Martin Lewis is the most-targeted, with Elon Musk, Richard Branson, Holly Willoughby, Phil Schofield and Piers Morgan also frequently impersonated. The figure appears to endorse a “new trading platform”, “wealth-building system” or “passive-income method”. The video routes to a sign-up page for an unregulated investment platform.
Red flags:
Real UK public figures do not endorse retail investment platforms in video ads. Martin Lewis has publicly and repeatedly said he has never endorsed any product, ever. Any video showing him doing so is, by his own statement, a deepfake. Similar disclaimers exist for the other commonly-impersonated figures.
Lip-sync mismatch. Watch the mouth carefully — deepfake lip movements often lag or lead the audio by 50-200ms. The mismatch is especially visible on long vowels and consonants like “p” / “b” / “m” where the lips should close completely.
Audio mismatch with visual context. The voice may sound slightly different from authentic recordings — intonation, pacing, breathing patterns. Compare to a known-real interview clip of the same person.
The video routes to an unregulated platform. When you check the platform name on the FCA Register, it’s absent or on the FCA Warning List. Run the platform through our Investment Pitch Analyser for a structured 8-pattern check.
Lighting and skin texture. Deepfake faces often have subtly “flat” lighting that doesn’t match the rest of the scene. Skin texture may look uniform or slightly waxy compared to natural photography.
Variant 2 — Deepfake CEO / executive Zoom or Teams call (business email compromise + video)
How it presents: A finance, payroll or M&A staff member receives a calendar invitation for a brief Zoom or Microsoft Teams call with the “CEO”, “CFO”, or external counsel. On the call, the executive’s face appears (deepfake), and they verbally instruct the staff member to transfer funds to a new supplier / acquisition target / legal escrow account before close of business. Often paired with a follow-up email from a lookalike domain.
Red flags:
The call is unusually short. Deepfakes degrade over time; scammers keep calls under 5 minutes. Real CEO instructions often warrant longer discussion.
The executive’s eye contact and blink rate look off. Real humans blink 15-20 times per minute. Deepfakes often blink less frequently or not at all in short clips. Watch for prolonged unbroken eye contact.
The face is centred and largely stationary. Deepfakes work best on a fairly still, well-lit, head-on shot. Real video calls have natural head movement, the executive looking off-camera, glancing at notes. A perfectly framed, motionless face is suspicious.
The connection “keeps cutting out” if asked to do something the deepfake can’t. Turning sideways, picking up an object, holding three fingers up — these gestures stress the deepfake system. A real executive will do them; the deepfake will lag, distort, or have a sudden “connection problem”.
Verification by callback is always available. Tell the executive you’ll call back via their saved internal number / Slack / Teams DM. Real executives welcome the verification. Deepfakes cannot survive a callback to a verified channel.
The transfer destination is new. Always cross-check against the standing supplier file. A new account / new sort code / “account changed” framing is the diagnostic BEC pattern with video now added.
Variant 3 — Deepfake romance / partner video (pig-butchering with video proof)
How it presents: A long-running romance or pig-butchering relationship (typically introduced via dating app, WhatsApp, Instagram, or LinkedIn) reaches the stage where the “partner” offers a video call to deepen trust. The deepfake video is short (30-90 seconds), lagged with a “bad connection” framing, often with the camera angle obscuring full facial movement.
Red flags:
Video calls are short, laggy, or always have a “bad connection”. This is intentional — limits the duration where the deepfake is most convincing.
The video angle is unusual. Side profile, low light, looking off-camera, or only the lower face visible. Real video calls usually have a head-on, well-lit view.
Real-time challenges fail or trigger “connection issues”. Ask the “partner” to turn their head fully sideways, then back. Real people can do this; current deepfakes degrade noticeably during fast lateral motion.
Hand-to-face interaction triggers artifacts. Ask them to touch their nose with the back of their hand, or hold three fingers up next to their face. The deepfake will distort around the hand.
The image static-frames perfectly with a still photo. Compare frames from the video call with photos they’ve sent. If they look almost identical (same lighting, same expression, same angle), the deepfake may be face-swapping onto a single source image.
The relationship has already involved money or is moving toward it. Run any investment / loan / business-opportunity offer through our Investment Pitch Analyser. Pig-butchering with deepfake video is one of the highest-loss-per-victim scam categories in UK 2026 reporting.
The detection signals that defeat current-generation deepfakes
Watch the blink rate. Count blinks over a 30-second clip. Real adults blink 7-10 times in 30 seconds. Deepfakes often blink less, with longer eye-open periods.
Watch the lip-sync precisely. Slow the video to half-speed if possible. Look at “p”, “b” and “m” sounds — lips should fully close. Look at long vowels (“ah”, “ee”) — mouth shape should match. Mismatches at sub-second precision are the strongest deepfake tell.
Look at hair and earring edges. Deepfake face-swap systems struggle with fine high-contrast detail at the boundary of the swapped region. Look for slight blurring, flickering, or stillness in hair strands and earring outlines while the head moves.
Watch lighting consistency. Real faces have natural lighting variation as the head moves. Deepfake faces often look uniformly lit or have lighting that doesn’t match the room (e.g. shadows on the body but not on the face).
Apply the side-profile / hand-gesture challenge. Ask the person to turn fully sideways for 5 seconds, then back. Ask them to hold three fingers up next to their face. Ask them to pick up an object and look at it. Each of these stresses current deepfake systems.
Verify via independent channel. Don’t trust a video call alone. Call the supposed person on a known phone number, or message them on a verified Slack / Teams / WhatsApp. Most deepfake operations are single-channel.
Reverse-image-search any video stills. Save a frame from the video and run it through Google Images or TinEye. Deepfake source images are often stock photos or images stolen from public profiles. A hit on an unrelated person’s profile is diagnostic.
For celebrity-endorsement videos: the figure’s real public statement is the test. Martin Lewis: never endorses anything. Elon Musk: doesn’t personally promote trading platforms. Branson: explicitly disclaims unauthorised endorsements. If the video shows them doing something they publicly say they never do, the video is fake.
If you’ve already lost money to a deepfake scam
UK bank transfer: Use the PSR Claim Wizard immediately. PSR Mandatory Reimbursement covers up to £85,000 within 5 working days for APP fraud, which includes deepfake-driven scams.
Crypto / foreign wire: Recovery is very limited. Report to Report Fraud on 0300 123 2040 and your bank. Specialised crypto-tracing firms exist but cost is high.
Business deepfake CEO loss: notify the bank’s corporate fraud team, the company’s insurer (most cyber-insurance policies cover BEC including deepfake-augmented BEC), and the genuine executive who was impersonated. Many BEC cases are partially recoverable if reported within 24 hours of the transfer.
Report the deepfake itself. Most major UK platforms have specific reporting for AI-generated impersonation content: Facebook / Instagram (Help Center > Report > AI-generated impersonation), TikTok (settings > report), YouTube (3-dot menu > Report > Misleading or deceptive practices).
Report to NCSC’s Suspicious Email Reporting Service (SERS) at report@phishing.gov.uk if the deepfake came via email or with a link.
If personal details / banking details were shared: consider CIFAS Protective Registration for 2 years of credit-file protection against follow-on identity fraud.