Google DeepMind’s SynthID technology embeds invisible watermarks in content made by artificial intelligence, helping people tell AI-generated media from real media.

The system works by altering AI-generated text, images, audio or video in ways humans cannot see or hear but computers can detect. The watermark survives editing such as cropping or compression.

SynthID matters because AI tools can now create realistic fake photos, voices and videos. Watermarks give platforms, publishers and fact-checkers a way to check where content came from. Media outlets and researchers can use detection tools to check whether a piece of content carries a SynthID mark. Google first described SynthID in 2023 and has since extended it to cover more media types.

The tool is being built into Google’s AI products, and other companies are adopting similar approaches. Regulators in several regions are pushing for clearer labelling of AI-made content.

Context: as generative AI spreads, watermarking is becoming a key defence against misinformation and deepfakes.