SECAI Core · phase 11 of 15 · noun
Deep fake
The use of artificial intelligence and machine learning to generate a highly-realistic video of a person. A fake video rendered using deep learning.
The Explain card
- Plain English
- A deep fake is a highly realistic fake video (or audio) of a person generated with deep learning. The face, voice, or both are synthesized so convincingly that viewers believe it is genuine.
- Example
- An employee receives a video message from the "CEO" announcing an urgent acquisition and instructing them to share confidential financials with an outside address.
- Why it matters
- Seeing and hearing are no longer proof. Defenders build verification steps for high-stakes requests, train staff to expect fakes, and consider content authentication tools.
- Hook
- A face you trust, wearing someone else's intent.
Word knowledge
How the word is built, where it came from, and what it sits beside in memory.
In a sentence
Investigators compared lip movements with the audio track before labeling the campaign video a deep fake.
Why these words
- deep Old English dēop, profound or extending far down links the phrase to deep learning
- fake English slang of uncertain origin, recorded in the late 18th century serves as the head noun and marks the item as fabricated
Where it came from
- Origin
- English compound formed from deep learning and fake, popularized through the Reddit account name deepfakes
- Entered the language
- 2017
- What changed
- It began as an online label, then entered journalism, computing, law, and security vocabulary; the closed spelling deepfake became standard.
How it is spelled
- Pattern
- The ee in deep is a common spelling for the long e sound.
- Pattern
- The final silent e in fake marks the preceding a as long.
- Breaks the pattern
- The supplied open form deep fake competes with the now-standard closed form deepfake.
Spelled like
- sleep
- keep
- make
- take
Broken into chunks
-
deep
- deep learning
- deep neural network
-
fake
- faker
- fakery
What it sits beside
Same subject
- voice cloning
- adversarial example
- data poisoning
- model inversion
Same shape
- false flag
- fake news
- voice clone
synthetic media terms
- deep fake
- voice clone
- synthetic image
AI security threats
- prompt injection
- model extraction
- data poisoning
Where it sits in the deck
Phase 11: Threat Landscape: Attack Vectors and Adversarial Techniques
With defenses named, learn what they must defend against — the full catalog of attack techniques targeting AI systems, their inputs, outputs, training pipelines, and supply chains.