Deepfake

Definition

A deepfake is AI-generated synthetic media — including video, audio, or images — designed to convincingly replicate or manipulate the appearance, voice, or actions of real individuals. Deepfakes are produced using deep learning techniques such as generative adversarial networks (GANs) and diffusion models, and have become increasingly accessible through consumer-grade tools. The term encompasses both visual and audio manipulation, and applies to media created for fraud, impersonation, disinformation, or non-consensual exploitation.

How It Relates to AI Threats

Deepfakes are central to threats across multiple domains. Within Information Integrity, they enable identity hijacking, synthetic media manipulation, and election interference through fabricated video or audio of public figures. Within Security & Cyber, deepfakes are used in business email compromise, CEO fraud, and social engineering attacks where attackers impersonate trusted individuals via video or voice to authorise fraudulent transactions.

Why It Occurs

• Generative AI models have become capable of producing highly realistic synthetic media from minimal source material
• Open-source tools and commercial platforms have lowered the barrier to creation
• Audio deepfakes require as little as a few seconds of sample voice to produce convincing clones
• Detection tools lag behind generation capabilities
• There is no universal authentication standard for verifying the provenance of media

Real-World Context

Deepfakes have been documented in financial fraud (Hong Kong CFO impersonation, UK energy CEO voice fraud), election interference (Slovakia 2023 election audio), identity impersonation (FBI deepfake warnings), non-consensual intimate imagery (Westfield High School), and institutional manipulation (Pikesville High School principal). The FBI has issued specific advisories regarding deepfake-enabled impersonation of government officials.