Trusted by teams building safer onboarding flows
SOC 2
iBetaSoon
Basic Deepfake
Detection Was Built
For Yesterday's
Attacks.
Fraud teams now face manipulated selfie sessions, not just photos and replays. Deloitte reports fintech deepfakes rose sharply in 2023, with AI-enabled U.S. fraud losses projected to reach $40B by 2027.
Stop KYC Fraud
Deepfake bypass attempts in onboarding have grown 3× YoY. Dataspike catches face swaps and replay attacks before a fraudulent account is opened.
Protect Live Calls
Voice-cloned executives and deepfaked video calls are now a documented attack vector in enterprise fraud. Real-time detection over WebRTC catches the attack before any action is authorized.
No Compliance Risk
SOC 2 Type II certified. GDPR-compliant. Helps your organization meet EU AI Act Article 50. On-prem and air-gapped deployment available. Full DPA on request.
Built For The Attacks You're
Already Seeing
Detection accuracy on face-swap attacks in KYC flows, including previously unseen models
Attack Vector
Attackers use AI face-swap tools to bypass video KYC checks — passing as legitimate customers to open accounts, take out loans, or launder funds. Face-swap bypass attempts have grown 3× year over year.
How Dataspike Stops it
On-device SDK runs invisibly inside your KYC flow. Detects face swaps, replay attacks, and digital masks in video. Image API checks ID photos and passport scans for AI manipulation. Nothing leaves the device for on-device deployments — zero UX friction.
- Passport & ID document tampering detection
- Inswapper, HeyGen, Hedra, StyleGAN and 30+ models
- Replay attack and overlay mask detection
- Works offline — nothing leaves the device
- 1–2 day SDK integration, no ML expertise required
See the result
in seconds
Built For The Two
Ways Attackers Strike
On-Device SDK
- Privacy is non-negotiable
- Device may be offline
- KYC / onboarding flow
- Proctoring & assessments
Cloud
- Live streaming or video calls
- Behavioral analysis across sessions
- Web3 wallet verification
- High-volume batch processing
32 AI models covered
Sora 2, Veo 3 and newer models have eliminated the visual artifacts that made deepfakes detectable by eye. Detection now requires infrastructure-level analysis — not human review. New models are added within days of release.
Not a black box —
a forensic result
Every detection returns a verdict, frame-by-frame analysis with Grad-CAM heatmaps, forensic flags, and an exportable PDF report. Your team knows exactly what was caught and why.
Exportable PDF report
Every detection generates a structured report with verdict, frame scores, heatmaps, forensic flags, and extracted metadata. Ready for compliance audit trails and regulatory review, with PDF export available.
Frame-by-frame analysis
Each video frame is independently scored. Top 3 highest-probability frames returned with timestamps. Temporal smoothing eliminates noise from single-frame artifacts.
Grad-CAM heatmaps
Visual explanation of exactly which facial regions drove the model's prediction. Red = highest model attention. Turns a probabilistic score into explainable evidence.
Forensic flags + metadata analysis
Structural and encoding metadata extracted via FFprobe and EXIF/XMP. Flags raised for missing camera data, AI generator signatures, and encoding inconsistencies.
Submit → Poll → Act
Simple API. No ML Expertise
Required. Standard REST
Upload your file
POST your video, image, or audio file to the API. Receive a job_id immediately — processing runs async on the backend.
POST /api/v4/deepfake/video/analyze
# returns job_id instantly
→ { status: "queued" }Check the result
Poll the job endpoint every 1–3 seconds until status becomes completed. Use exponential backoff to stay within rate limits.
GET /api/v4/deepfake/job/{id}
# poll until completed
→ { status: "completed" }Read score & verdict
Use the score (0–1) and verdict to block, flag, or escalate in your existing logic. Sandbox available for testing before production.
{
score: 0.985,
verdict: "deepfake_likely"
}Start With Testing.
Scale when risk or volume proves it. Add Dataspike Deepfake Detection as a standalone module or layer it into your existing KYC stack — run it on your dataset, or embed it through API, SDK, or OEM licensing.
Test on your fraud cases
Benchmark against labeled good-user sessions, failed reviews, suspicious cases, and known fraud.
Test deepfake evaluation in sandbox, move to pay-per-check in production, and switch to custom pricing as your volume grows.
Pay as you go
Volume pricing
Get custom pricing for higher verification volumes, advanced deployment needs, or broader KYC and fraud-control bundles.
Talk to salesWhy Dataspike for deepfake detection?
What is deepfake detection and how does it work?
Deepfake detection analyzes statistical artifacts left by generative AI models in video, image, and audio. The engine returns a 0–1 confidence score and a human-readable verdict — not just a database of known fakes.
What is the difference between liveness and deepfake detection?
Liveness checks that a real person is present in the session. Deepfake detection checks whether the media itself has been AI-manipulated. They catch different attacks — used together they cover the full threat surface.
Can it detect deepfakes from models it hasn't seen before?
Yes. The detection engine analyzes statistical artifacts left by generative architectures — not a fixed database of known fakes. When a new model is released, coverage is typically added within days.
What types of AI-generated fraud can Dataspike detect?
Face swaps, replay attacks, synthetic identities, AI-generated voices, talking avatars, digital masks, and Instagram-style filters. 30+ generative models covered — updated weekly.
How do we integrate deepfake detection?
How fast is detection in a live call or KYC session?
The cloud API returns a result in 1–2ms per frame. The on-device SDK processes at 3–20ms per frame with no network dependency. Detection runs invisibly alongside the session with no impact on user experience.
When should I use the SDK vs the cloud API?
On-device SDK is best when privacy is non-negotiable or the device may be offline — KYC flows, proctoring, Web3 verification. The cloud API is better for live streaming and video conferencing where real-time behavioral analysis adds accuracy.
How long does integration take?
Basic integration takes 1–2 days for either the SDK or the API. Both are well-documented with no ML expertise required. The SDK embeds directly into your mobile or web app; the API connects via standard HTTP/HTTPS from your backend.
What formats and modalities are supported?
Video: MP4, WebM (REST API) or WebRTC via Pipecat and LiveKit plugins (streaming). Image: JPG, PNG, WebP. Audio: MP3, WAV, M4A. SDK: iOS, Android, WebAssembly.
Compliance & Security
Does Dataspike store the video or audio I send?
Media submitted for analysis is not stored permanently. Verification results and metadata are automatically deleted after 180 days. For on-device SDK deployments, nothing leaves the device at all. Enterprise clients can negotiate custom retention terms. Full DPA available on request.
Is Dataspike GDPR and EU AI Act compliant?
Dataspike is SOC 2 Type II certified and GDPR-compatible. On EU AI Act — Article 50 requires organizations to detect and disclose deepfakes. Dataspike is the detection layer that helps your organization meet that obligation. Data is automatically deleted after 180 days.
KYT Crypto Feature
Monitor crypto wallets and transactions for risk signals.
Deepfake Detection
Detect manipulated media across image, video, audio, and document inputs.
Compliance Suite
Screen users, monitor risk, and keep review evidence in one workflow.
