Deepfake

A deepfake is synthetic media — image, audio, or video — that depicts a real person doing or saying something they did not actually do, produced by AI generation or face/voice swap.

By 2026 deepfake quality is high enough to fool casual viewers on short clips. Most major social platforms (Meta, X, YouTube, TikTok, Pinterest) require deepfake labelling, and EU AI Act + US state laws criminalise non-consensual intimate deepfakes and election-interference deepfakes. The defence stack: provenance manifests (C2PA), inaudible watermarks (SynthID, Resemble PerTh, ElevenLabs voice watermarks), and deepfake detectors (Microsoft Video Authenticator successors, Intel FakeCatcher). For publishers and platforms, the standard is: watermark or label any AI-generated media before publishing, and detect-then-label inbound content where possible.

Common mistakes

• Treating watermarks as proof of authenticity — they prove origin, not safety.
• Skipping consent collection on voice / face cloning — legal exposure is rising globally.

FAQ

Related terms

• AI watermarking — AI watermarking embeds invisible-to-humans signals in model output (text, image, audio, video) so the content can later be detected as AI-generated.
• Content provenance — Content provenance is cryptographic metadata attached to media that records how it was created, by whom or what model, and what edits it has been through.
• Multimodal model — A multimodal model accepts more than one input type — text plus images, audio, or video — and reasons across them in a single forward pass.

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/deepfake.md.