Glossary of AI Detection Terms

SynthID

An invisible digital watermarking technology developed by Google DeepMind. It embeds a cryptographic signal directly into the pixels of an image or the waveforms of audio at the moment of generation, allowing compatible detectors to identify it even after cropping or compression.

C2PA (Coalition for Content Provenance and Authenticity)

An open technical standard providing publishers, creators, and consumers the ability to trace the origin of different types of media. Unlike SynthID, C2PA relies on cryptographic metadata securely attached to the file, rather than altering the image pixels.

Content Credentials

The consumer-facing term for C2PA metadata. Often visible as a small "CR" pin icon on an image, it allows users to click and inspect the file's history, including which camera captured it or which AI generated it.

Invisible Watermark

A marker covertly embedded within a digital signal (like an image or audio file) that is imperceptible to human senses but can be detected by specialized algorithms.

Adversarial Attack

A technique designed to trick or fool machine learning models. In the context of SynthID, this involves systematically applying destructive edits (noise, filters) to an image in an attempt to completely erase the watermark without ruining the image quality.