Highlights
3 years in a row named a Leader
A Leader in the 2026 Gartner® Magic Quadrant™ for Identity Verification
Download the report
Validated under real-world attack scenarios
First to achieve iBeta Level 3 on iOS and Android
Learn more
Introducing GovFaceMatch
The first identity solution to match biometrics against DMV records
Read more
How we protect data
Privacy is the architecture
Explore the Privacy Hub
Deepsight detects AI-generated faces and blocks injected or tampered camera feeds in real time, with no added friction for genuine users.
The challenge
Identity verification assumes the face on camera is a real, live human. That assumption is breaking: AI can generate a convincing deepfake of anyone, and attackers inject that synthetic video straight into the verification stream, bypassing the camera entirely.
Most people struggle to tell a deepfake from a real face, and standard liveness catches masks and replays rather than AI-generated faces or injected media. To both, a good deepfake looks like a genuine selfie.
of people can't distinguish between a deepfake and a real image or video
Diel et al., Computers in Human Behavior Reports
forecast AI-enabled fraud losses by 2027
Deloitte
the cost of producing a convincing deepfake
FinCEN
YoY growth in fintech deepfake incidents
Deloitte
How it works
All checks run invisibly in the background, introducing real-time security monitoring and fraud analysis on every attack angle that a fraudster could use to inject deepfakes and spoofs into the verification process.
Why Incode
Deepsight catches what liveness checks and human reviewers miss, without adding a single step for genuine users.
Accuracy
Independent study conducted by Purdue University identified Incode as the most accurate deepfake detector:
68x
better false-positive
rate in Identity Verification than the next-best commercial technology
2.5x
lower false-acceptance
rate across all deepfake samples
Security
24,360
fraudulent sessions caught in shadow mode
missed by standard liveness and manual review
Device and camera integrity stops the virtual cameras, emulators, and injected feeds that pure deepfake detectors ignore.
Speed
<100ms
passive processing
transparent to the user, with no extra capture steps
Genuine users complete verification with no added wait.
Scalability
7.1B+
trust checks run on the Incode platform
real-world attack data feeding detection models
Incode's in-house R&D team and Fraud Lab continuously train detection models against the latest deepfake techniques.
Use cases
Wherever a face is on camera, Deepsight verifies a real, live human is behind it.
Block deepfaked and injected captures inside identity verification, before a fraudulent account opens.
Screen image-based authentication attempts for AI-generated faces at login and step-up.
Stop real-time deepfakes in interviews, helpdesk calls, and employee verification.
Catch the deepfakes built to bypass age assurance.
Run deepfake checks on media captured by your existing IDV or authentication tool.
Methods
Deepsight's detection layers run inside Incode flows or standalone on media captured elsewhere.
AI-generated documents are the next frontier of identity fraud
Generative AI creates convincing synthetic documents as easily as convincing fake faces. Deepsight for Documents protects the document layer, catching forged IDs, passports, and supporting documents that traditional verification tools miss.
Its AI forgery detection feature identifies documents created or altered by generative AI tools by detecting visual artifacts, font inconsistencies, and layout anomalies invisible to the human eye.
10x
increase in AI-generated document fraud over two years
8.8x
more fraud caught by Deepsight for Documents than document-based checks alone
25%
of identity fraud attempts are now AI-assisted, estimated to reach 50% by the end of the year
Verified proof
10x
fewer false positives than expert human labelers
8 of 10
top U.S. banks choose Incode
1.4M
shadow-mode sessions analyzed
What customers say
“Deepfakes are rewriting the rules of fraud. Our work with Incode brings their Deepsight deepfake detection directly into Experian's identity and fraud solutions, giving our clients market-leading protection and keeping them one step ahead of AI-powered attacks.”
Keir Breitenfeld · Senior Vice President, Fraud and Identity, Experian
Deepfake detection is the process of identifying AI-generated synthetic media (altered videos, images, or audio) used to impersonate real people. In identity verification, deepfake detection software prevents fraudsters from using synthetic media to bypass biometric checks.
AI deepfake detectors analyze visual signals, pixel-level artifacts, and temporal inconsistencies that are absent or distorted in synthetic media. Incode's Deepsight runs three layers together: multi-modal AI across video, motion, and depth, device and camera integrity that blocks injected feeds, and behavioral signals that catch fraud farms, all passively and in under 100ms.
A video injection attack occurs when a fraudster routes a synthetic or pre-recorded video feed into an identity verification system through a virtual camera driver, bypassing liveness checks that only analyze camera input. Deepsight detects injection attacks at the signal level as well as the visual level.
No. Standard liveness catches masks and replays rather than AI-generated faces or injected feeds, so a video selfie won't protect you against a good deepfake. Deepsight is the layer that does.
Deepsight is Incode's proprietary deepfake and liveness detection engine. It uses a multi-layer AI model to identify AI-generated faces, video injection attacks, and presentation attacks in real time, validated by Purdue University as the most accurate system in its class.
Purdue University's Machine Learning Lab independently validated Deepsight as the most accurate deepfake detector: the lowest false-acceptance rate of the systems tested, 2.5x lower than the next-best commercial technology across all deepfake samples, and a 68x better false-positive rate in Identity Verification. It operates in milliseconds, with 10x fewer false positives than expert human labelers.
Yes. Across 1.4 million shadow-mode sessions, Deepsight caught 24,360 fraudulent sessions (1.74%) that standard liveness and manual review missed, and deepfake-enabled account openings, takeovers, and a $25 million deepfake CFO transfer are documented.
What's next
See how Deepsight stops deepfakes and injection attacks without adding friction.
Answers come from across incode.com. For the full explainer, ask anything.
