Deepfake: A deepfake is synthetic audio, video, or imagery, generated or altered by AI, that convincingly depicts a real person saying or doing something they never did.
Voice cloning tools can reproduce a person's voice from a short sample. Video models can swap faces, lip-sync speech, or generate a person from scratch. Image generators can place someone in a scene that never happened. Quality has reached the point where casual inspection is not enough, and the tools are cheap and widely available.
The workplace risk is fraud and impersonation. There have been widely reported cases of finance staff wiring money after video calls with what appeared to be their executives, and of employees receiving urgent voice messages from a cloned CEO. Deepfakes are also used to harass individuals, fake endorsements, and manufacture evidence. Legitimate uses exist too: synthetic presenters for training videos, dubbing, and accessibility, when the person has consented.
Defense is procedural. Any request to move money, change payment details, share credentials, or bypass a control gets verified through a separate channel you initiate, such as calling a known number. Agree on a team code word for urgent requests. Assume that a voice or a face on a call proves nothing by itself. Detection tools exist but are unreliable, so do not build your process around them.
Example at work
A controller receives a video call from someone who looks and sounds like the CFO, asking for an urgent vendor payment before a deadline. Company policy requires callback verification for any new payment instruction. She hangs up, calls the CFO's known number, and learns there was no such request. The policy, not her eye for fakes, caught it.
Why it matters
Deepfakes turn "I saw it with my own eyes" into an unreliable standard. Every team that handles money, credentials, or sensitive decisions needs a verification step that does not depend on recognizing a face or a voice.
Related terms
- AI watermarkingAI watermarking embeds a hidden or attached signal in AI-generated content, such as an invisible pattern in an image or metadata in a file, so that its origin can be identified later. It helps with provenance but is far from foolproof.
- AI detectorAn AI detector is a tool that tries to estimate whether a piece of text, an image, or audio was generated by AI. Detectors for text in particular are unreliable, with false positives and easy evasion, and should not be used as sole evidence.
- Text-to-speech (TTS)Text-to-speech (TTS) is AI that converts written text into natural-sounding spoken audio. Modern systems produce voices with realistic pacing and emotion, and some can clone a specific person's voice from a short sample.
- Generative AIGenerative AI is a class of AI models that create new content, including text, images, code, audio, and video, in response to a prompt. Chat assistants like ChatGPT, Claude, and Gemini and image tools like Midjourney are generative AI.
- AI safetyAI safety is the field concerned with preventing AI systems from causing harm, from everyday failures like confident errors and biased outputs to misuse by bad actors and risks from highly capable future systems.