Neural Defend
Start Here

Industries Marketplaces & consumer apps

Your platform trusts a face and a photo. Both can now be generated.

Sellers, couriers, hosts and dates prove who they are with a selfie; buyers prove damage with a photo; owners win back an account with a video. NeuroVerify checks the face at onboarding, re-verification and recovery, and NeuroDetect gives your trust and safety team forensic findings on refund evidence and reported media.

See where it hitsTest it on your own cases

The problem, in numbers

Regulators, platforms and merchants are counting the same thing.

  • $2.1bn

    reported lost in 2025 to scams that started on social media, eight times the 2020 figure. Nearly 60% of people who lost money to a romance scam said it began on a social platform.

    US FTC Data Spotlight, 27 Apr 2026
  • 134m

    scam ads Meta says it removed in 2025 by early December. It now matches faces in suspect ads against public figures’ profiles, and says this more than doubled the celeb-bait ads it caught in testing.

    Meta, 3 Dec 2025
  • 9%

    of UK people aged 16+ are confident they can spot a deepfake. 43% believe they saw at least one in the previous six months.

    Ofcom, 23 Jul 2024
  • 195 yuan

    refunded to a buyer who faked a video of dead crabs with AI. In another case the platform’s service centre said it could not tell whether the refund photos were AI-generated.

    China Daily, 12 Dec 2025

Where deepfakes hit

Six workflows where a fake buys access, money or reach.

Each one ends with a person or a rule accepting media they cannot inspect in person. That acceptance is what the attacker is buying.

  • Marketplaces · rentals

    Seller and host onboarding

    A trader account is opened with a stolen ID and a face swap of that ID’s owner, streamed into the selfie step through a virtual camera. The face match passes, because the face is the right one.

  • Gig · delivery · ride-hailing

    Worker re-verification and account sharing

    A verified worker rents the account to someone who could not pass screening. The renter answers each surprise selfie with a replay or a live face swap of the owner, so the match succeeds every shift.

  • E-commerce · quick commerce

    Refund and dispute evidence

    The buyer keeps the item and submits a generated defect: a cracked screen, mouldy fruit, a dead crab. Where policy refunds without a return, that image is the only evidence.

  • Dating · social

    Romance profiles

    A profile is built from generated photos, and a video call with the same synthetic face “proves” the person is real. The FBI lists romance schemes among the uses of AI-generated profile images.

  • Social · creator platforms · ads

    Creator and celebrity impersonation

    A known face and voice are cloned into an ad or a lookalike account selling a giveaway, a course or an investment. Reports reach trust and safety as clips and screenshots, with no original beside them.

  • Every consumer app

    Account takeover via recovery

    The attacker claims a lost phone and submits a selfie or video built from the owner’s public photos. One approved reset hands over a seller’s payouts, a creator’s audience or a rider’s saved cards.

Why your checks pass it

Your controls ask who it is. None asks whether it is real.

A face match is built to say yes to the right face. A deepfake of the right face is exactly what it was built to accept.

CheckThe question it answersWhat gets through

Selfie-to-ID face match

Is this the person on the document?

A face swap of the document’s owner. The match succeeds because the faces are the same.

Liveness and “verified” badges

Is a person in front of the camera?

Video injected through a virtual camera, and face swaps that blink and turn on cue.

Periodic re-verification selfie

Is the account owner working right now?

A renter replaying or face-swapping the owner. The check confirms whose face it is, not how it was captured.

Duplicate-face and reverse image search

Have we seen this face or photo before?

A face or a defect generated for this one submission. There is nothing earlier to match.

Device and behaviour signals

Does this session act like the owner?

A renter on the owner’s phone. A brand-new seller with a clean device and no history to compare.

Refund agent or moderator

Does this look right?

Media generated to look right, judged at queue speed. Only 9% of UK adults told Ofcom they feel confident spotting a deepfake.

NeuroVerify

Is this face, video or voice real?

One risk score with six signals, including Live Presence Verification and Face Manipulation Detection, on the selfies you already collect.

NeuroDetect

Was this file edited or generated?

Region-level findings, heatmaps and a forensic report on refund photos, reported clips and ad creatives.

For the product rows, the last column is what the check adds.

One platform, six doors

Faces go to NeuroVerify. Files go to NeuroDetect.

The split follows the media. A selfie at a checkpoint needs an answer in milliseconds. A refund photo or a reported ad needs an analyst with a report.

Trust & safety queueOne platform, six doors
  1. Seller onboarding

    Selfie video, new trader

    NeuroVerify

    Face manipulation

    Hold. Trader cannot list.

  2. Courier re-check

    Surprise selfie, shift start

    NeuroVerify

    Not a live capture

    Block going online. Review.

  3. Refund claim

    Photo, “cracked on arrival”

    NeuroDetect

    Generated region

    No auto-refund. Agent review.

  4. Dating verification

    Video selfie vs profile photos

    NeuroVerify

    No manipulation found

    Badge granted.

  5. Reported ad

    Clip using a creator’s face

    NeuroDetect

    Face manipulation

    Escalate to ads integrity.

  6. Account recovery

    Selfie, “lost my phone”

    NeuroVerify

    AI-generated face

    Stop reset. Call-back path.

Illustrative queue. The finding is evidence; the hold, block, refund or badge is your platform’s policy.

Regulation and guidance

Lawmakers now expect platforms to know who is selling and advertising.

A plain summary of what each text says, not legal advice. Neural Defend supplies evidence. It does not make a platform compliant.

  • EU · Digital Services Act, Art. 30

    Know the trader before they sell

    Online marketplaces must obtain a trader’s contact and payment details and a copy of an identification document before the trader can offer goods to EU consumers. NeuroVerify does not read the document. It checks the selfie or video that ties a person to it.

    Regulation (EU) 2022/2065
  • EU · Digital Services Act, Arts. 34–35

    Systemic risk at the largest platforms

    Very large platforms assess systemic risks at least once a year and before launching features likely to have a critical impact on them. Article 35(1)(k) lists prominent labelling of generated or manipulated media that resembles real people as one mitigation measure.

    Regulation (EU) 2022/2065
  • EU · Digital Services Act, Art. 22

    Trusted flaggers go first

    Notices from designated trusted flaggers, acting within their area of expertise, must be processed with priority and decided without undue delay. A forensic finding helps a reviewer decide fast without deciding blind.

    Regulation (EU) 2022/2065
  • UK · Online Safety Act

    Duties on fraudulent paid ads

    Category 1 and 2A services carry fraudulent-advertising duties. Ofcom published draft codes on 10 Jul 2026 covering ad moderation, advertiser account integrity and complaints. Consultation closed on 2 Oct 2026; a final statement is due by mid-2027 at the latest.

    Ofcom consultation
  • US · Federal Trade Commission

    Platforms and impersonation ads

    On 24 Sep 2026 the FTC asked for comment on whether to update its Impersonation Rule to address platforms, including whether they should be required to stop impersonation ads from reaching consumers. It is a request for comment, not a rule.

    FTC press release

Integration

One call behind every selfie. One workspace for every report.

Step 1

Add NeuroVerify to the selfies you take

Onboarding, surprise re-checks and recovery. Native SDKs for Android, iOS and Flutter.

Step 2

Set a threshold per door

A courier re-check and a top seller’s recovery carry different risk. Your policy sets each one.

Step 3

Send flagged files to NeuroDetect

Refund evidence, reported profiles and ad creatives open in a case workspace for your analysts.

Step 4

Route on the result

Hold, ask for a new capture, or escalate. Keep the report for the user’s appeal.

SDK-first integration

One API.
Every stack.

NeuroVerify inside your seller, courier and dating apps: native SDKs for Android, iOS and Flutter, plus Python, TypeScript, Go and Java for your backend. Two lines to integrate; live in under 24 hours.

Sign in for API access
$ npm install @neuraldefend/sdk

import { NeuroVerifyClient } from "@neuraldefend/sdk"
const result = await new NeuroVerifyClient().detectImage("selfie.jpg")

Mobile SDKs

Native capture apps
  • KAndroid (Kotlin)
  • SiOS (Swift)
  • FFlutter

Backend SDKs

Server-side checks
  • PyPython
  • TSTypeScript / Node.js
  • GoGo
  • JJava

REST API

For everything else

Already have capture? Send the image to POST /detect/image and get every signal back in one JSON response.

AI agents

MCP server

Give approved AI assistants bounded image and video authenticity tools, with API keys kept outside prompts.

Two lines to integrate<200 ms p95Live in under 24 hours

NeuroDetect runs as a web app, in your private cloud or VPC, as a secure container, on-premises or air-gapped. ISO/IEC 27001 · 27701 · 42001 · SOC 2 Type II.

Knowledge centre

Read more on marketplace and consumer-app fraud.

View all insights

Sources

Read the evidence yourself.

Third-party publications. None of them evaluates or endorses Neural Defend. Platform figures are as reported by each company.

Does NeuroVerify check seller ID documents?

No. NeuroVerify analyses face images, video and audio. It sits beside your document check and answers a different question: is the face in the selfie real, live and unmanipulated. Digital onboarding

We already run selfie re-checks on couriers. What does this add?

Your re-check confirms the face belongs to the account owner. A replay or a face swap of the owner passes that test by design. NeuroVerify scores the same selfie for signs it was not captured live or was manipulated.

Can a score suspend a seller or a worker on its own?

It should not. NeuroVerify returns a risk score with the signals behind it, and your policy decides. For actions that cost someone their income, route high scores to a reviewer and keep the evidence for the appeal.

Will honest users notice?

The check runs on the selfie you already collect, so there is no new step. Eligible image requests return in under 200 ms at p95 under defined test conditions.

How do moderators use NeuroDetect on reported content?

A reported clip, profile photo or ad creative opens in a case workspace with heatmaps, frame and region views and a plain-language summary linked to the evidence. The moderator makes the call; the forensic report records how. NeuroDetect

Where does user media go?

Managed cloud, your own VPC, a secure container, on-premises or air-gapped for NeuroDetect. Neural Defend holds ISO/IEC 27001, 27701 and 42001 certifications and a SOC 2 Type II report. Product data handling

See it on your own cases

Run last quarter’s banned accounts through it.

Bring onboarding selfies, re-check captures, refund photos and reported media, including cases you already know were fraud. See what NeuroVerify and NeuroDetect find, and what lands in the report.

Book an evaluationExplore NeuroVerifyExplore NeuroDetect