What Is a Deepfake

A deepfake refers to technology - and its output - that uses deep learning to convincingly synthesize or replace a person's face or voice. Rapid advances in generative AI technologies such as GANs (Generative Adversarial Networks) and diffusion models have made it possible to create high-quality forged media without specialized expertise.

Originally developed in the context of movie visual effects and academic research, the technology is now being weaponized for social engineering, disinformation campaigns, and fraud. The barrier to creating deepfakes continues to drop, making this an increasingly urgent security concern.

Abuse Tactics and Damage Patterns

Deepfake-powered attacks dramatically amplify the persuasiveness of traditional social engineering. The main abuse patterns are as follows.

Evolved Business Email Compromise: Synthesizing an executive's voice in real time to instruct urgent wire transfers over the phone. In 2019, a UK energy company was defrauded of $243,000 through a deepfake voice call impersonating the CEO.
Disinformation and Fake News: Fabricating videos of politicians or public figures making statements they never made to manipulate public opinion or financial markets. During elections, deepfake videos can spread rapidly on social media.
Identity Fraud: Using deepfake faces to bypass video-based identity verification (eKYC). Financial institutions and cryptocurrency exchanges that rely on selfie verification are at risk.
Harassment and Extortion: Creating non-consensual intimate imagery of individuals for harassment or blackmail. Victims suffer severe psychological harm, and once distributed online, removal is extremely difficult.

Detection Technologies and How to Spot Deepfakes

Deepfake detection is an ongoing arms race with generation technology, but several effective approaches currently exist.

Technical Detection Methods

Biological Signal Analysis: Detecting subtle pulse variations (rPPG) and blink patterns present in real video. Synthetic video lacks or exhibits unnatural biological signals.
Frequency Domain Analysis: GAN-generated images contain characteristic artifacts in the frequency domain that are invisible to the naked eye but detectable by algorithms.
Temporal Consistency Checks: Analyzing frame-to-frame consistency in video. Deepfakes may exhibit unnatural flickering or inconsistencies in lighting and shadows.

Visual Clues for Human Detection

Unnatural boundaries between face and hair/ears
Inconsistent lighting direction between face and background
Teeth or eye reflections that appear blurred or distorted
Lip movements that don't perfectly sync with audio

However, the latest generation models are rapidly eliminating these visual artifacts. Relying solely on human observation is becoming insufficient - technical detection tools are increasingly necessary.

Defense Strategies for Individuals and Organizations

Defending against deepfake threats requires combining technical and operational measures.

Individual Measures

Limit Public Exposure of Photos and Videos on Social Media: The more facial data available publicly, the higher the risk of high-quality deepfake generation.
Verify Suspicious Communications: When receiving calls or video messages claiming to be from executives or family members requesting money, verify through a separate channel (call back on a known number).
Establish Code Words: Agree on a secret code word with family members for emergency situations to verify identity during suspicious calls.

Organizational Measures

Multi-Person Approval for Financial Transactions: Never process large transfers based on a single phone call or video instruction. Require approval from multiple authorized personnel.
Deepfake Awareness Training: Include deepfake scenarios in security awareness training so employees can recognize potential attacks.
Content Authentication: Adopt standards like C2PA (Coalition for Content Provenance and Authenticity) to embed provenance information in images and videos, enabling verification of authenticity.

To learn more about this topic, see How to Spot Deepfakes: Protecting Yourself from Fake Videos and Audio.

Common Misconceptions

Only experts can create deepfakes: With the proliferation of open-source tools and cloud services, it is now possible to perform face swaps and voice synthesis in minutes without technical expertise. Even smartphone apps can generate basic deepfakes.
You can always spot a deepfake by watching carefully: State-of-the-art generation models produce deepfakes that are difficult for humans to distinguish visually. Low-resolution video and short audio clips are particularly challenging - even experts may struggle to make accurate judgments. Technical detection tools are essential.

Deepfake