Injection attack
Also called digital injection attack, video injection attack, camera injection
Feeding forged video or audio straight into an identity check or call, bypassing the camera or microphone. Not the same thing as prompt injection.
When an app checks your face, it trusts that the video came from your camera. An injection attack breaks that trust. The attacker replaces the camera feed with a file or a live AI-generated stream, so the app receives a clean, well-lit, perfectly matching face that no camera ever saw.
Generative AI is what made this practical. Tools that produce a convincing, moving face are now effective and widely available, and the standard for identity checks has been rewritten to assume attackers have them.
The defences are different from those for a photo held up to a lens. A provider has to check that the media came from a real capture device and look for signs of forgery in it. An organisation relying on the check should add something a forged picture can't carry, on a channel the applicant doesn't control.
For example
A fraudster uses software that poses as a webcam to play an AI-made face into a bank's account-opening check.
Put it to use
Where you'll meet injection attack in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.