The Deepfake Fraud Glossary

Clear, citable definitions of the terminology behind AI-generated fraud and its detection, written for fraud, risk, and compliance teams in financial services.

Browse terms

Deepfake

A deepfake is audio, image, or video content generated or manipulated by artificial intelligence to convincingly imitate a real person's face, voice, or actions. Modern deepfakes are produced by generative models such as diffusion networks, GANs, and neural voice synthesizers, and are frequently indistinguishable from authentic media to the human eye and ear.

Read definition

Synthetic Media

Synthetic media is any audio, image, video, or text content that has been wholly generated or substantially altered by AI systems. Deepfakes are the impersonation-focused subset of synthetic media; the broader category also covers AI-generated identity documents, fabricated screenshots, and machine-written scripts used in fraud campaigns.

Read definition

Voice Cloning

Voice cloning (also called voice spoofing) is the use of AI speech synthesis to replicate a specific person's voice, often from only a few seconds of sample audio taken from voicemail greetings, earnings calls, or social media. The cloned voice can then say anything the attacker types, in real time.

Read definition

Face Swap

A face swap is a deepfake technique that replaces one person's face with another's in an image or video while preserving expressions, lighting, and head movement. Real-time face-swap tools can now run inside a live webcam feed, letting an attacker present someone else's face during a video call.

Read definition

Deepfake Detection

Deepfake detection is the automated analysis of media to determine whether it was generated or manipulated by AI. Detection models examine artifacts that generative systems leave behind — spectral patterns in synthetic speech, frequency-domain anomalies in images, temporal inconsistencies in video — and return a confidence score rather than relying on human inspection.

Read definition

Audio Deepfake Detection

Audio deepfake detection is the analysis of speech audio to identify signs of synthetic generation or voice conversion. Unlike speaker verification, which asks "is this the enrolled person?", audio deepfake detection asks "was this audio produced by a machine?" — examining spectral, prosodic, and phase artifacts characteristic of text-to-speech and voice-cloning models.

Read definition

Liveness Detection

Liveness detection verifies that biometric input comes from a physically present, live human rather than a photo, recording, mask, or digital replay. Active liveness asks the user to perform actions (blink, turn); passive liveness analyzes texture, depth, and micro-movement without user interaction. Liveness spoofing is the practice of defeating these checks with deepfakes or injection attacks.

Read definition

Presentation Attack

A presentation attack is an attempt to defeat biometric verification by presenting falsified evidence to the sensor: a printed photo, a screen replay, a silicone mask, or a deepfake displayed to the camera. It is the "in front of the camera" attack class, standardized in ISO/IEC 30107 alongside presentation attack detection (PAD).

Read definition

Injection Attack

An injection attack bypasses the physical camera or microphone and feeds pre-made or real-time synthetic media directly into the verification data stream, using virtual cameras, tampered apps, or intercepted API traffic. Unlike a presentation attack, nothing is shown to a sensor — the fake enters the pipeline as data.

Read definition

Synthetic Identity Fraud

Synthetic identity fraud is the creation of a new, fictitious identity by combining real elements (often a valid SSN or national ID number) with fabricated names, documents, and increasingly AI-generated faces. Unlike classic identity theft, there is no real victim to report the fraud — the identity belongs to no one.

Read definition

Deepfake Vishing

Deepfake vishing is voice phishing (vishing) enhanced with AI voice cloning: the attacker calls a victim — a customer, an employee, or a call-center agent — using a synthesized voice that imitates a trusted person such as a family member, a bank representative, or an executive. Combined with caller-ID spoofing, the call is nearly indistinguishable from a legitimate one.

Read definition

Executive Impersonation (CEO Fraud)

Executive impersonation, often called CEO fraud, is the use of deepfake voice or video to pose as senior leadership and pressure employees into urgent, confidential actions — typically wire transfers or credential disclosure. It is the deepfake evolution of business email compromise (BEC), moving the deception from a spoofed inbox to a live call.

Read definition

KYC Deepfake Fraud

KYC deepfake fraud is the use of AI-generated or AI-manipulated media to defeat Know Your Customer verification: forged or synthesized identity documents, AI-generated portrait photos, face-swapped selfies, and deepfake video streams in video-KYC interviews. It targets the exact controls regulators require for account opening.

Read definition

Account Takeover (ATO)

Account takeover (ATO) is fraud in which an attacker gains control of a legitimate customer's existing account — changing credentials, adding payees or devices, and moving funds out. Modern ATO increasingly runs through the call center, where a cloned voice plus breached personal data convinces an agent to reset the account's defenses.

Read definition

Authorized Push Payment (APP) Fraud

Authorized push payment (APP) fraud is a scam in which the victim is manipulated into authorizing a payment themselves — to an account controlled by the fraudster. Because the account holder approves the transfer, it bypasses controls built for unauthorized transactions; deepfaked voices and videos of family members, bank staff, or executives are now the manipulation tool of choice.

Read definition

Media Forensics

Media forensics is the scientific analysis of audio, images, and video to establish authenticity, provenance, and evidence of manipulation. Techniques include metadata analysis, sensor-noise fingerprinting, compression-history examination, frequency-domain analysis, and — against generative AI — detection of the statistical artifacts synthesis models leave in their output.

Read definition

Content Provenance (C2PA)

Content provenance is the verifiable record of a media file's origin and edit history. The C2PA standard (Coalition for Content Provenance and Authenticity) implements it as cryptographically signed "Content Credentials" embedded at capture or creation, letting downstream consumers verify who produced a file and what was changed.

Read definition

Call Center Authentication

Call center authentication is the process of verifying a caller's identity before allowing account access or changes, traditionally through knowledge-based authentication (KBA), one-time passcodes, caller-ID matching, or voice biometrics. It is the control that decides whether the person on the line is the customer.

Read definition

Real-Time Voice Authentication

Real-time voice authentication is the continuous verification of a caller's voice during a live conversation, rather than a single check at call start. Modern implementations pair speaker verification (is this the enrolled voice?) with synthetic-speech detection (is this voice machine-generated?) so that both impostors and deepfakes are caught mid-call.

Read definition

Detection Confidence Score

A detection confidence score is a numeric output expressing how strongly analyzed media exhibits characteristics of AI generation or manipulation. Rather than a binary real/fake verdict, the score quantifies evidence strength, allowing consumers of the score to choose their own decision thresholds per use case.

Read definition

Looking for buying answers instead?

Our FAQ covers questions like which deepfake detection fits banks, how the API works, and how to detect deepfakes in KYC.