The industry is settling on two complementary tools to label AI-generated media: SynthID, Google's invisible watermark built into the content itself, and C2PA "content credentials", which are signed metadata recording where a file came from. Google announced at I/O 2026 that SynthID and C2PA checks are coming to Chrome and Search, and OpenAI has said it will embed SynthID in ChatGPT images and join the C2PA steering committee. Much of the consumer-facing rollout is still "coming months", not live.
How the two approaches differ
- C2PA content credentials attach a signed record of a file's origin and edits. If anyone tampers with the file, the signature breaks. But metadata is easy to strip: a screenshot or re-upload can remove it.
- SynthID changes the pixels or audio themselves in a way people can't see, so it can survive where metadata gets stripped.
Used together, one gives you a verifiable history and the other survives copying.
The limits
- Coverage: SynthID only marks content from models that use it. An image from a tool that doesn't watermark has nothing to detect.
- Removal: researchers have shown that some watermark schemes can be weakened or removed, for example through fine-tuning.
- No label isn't proof. Unmarked content isn't necessarily real.
What businesses should do now
- Verify before you act on "urgent" voice notes or videos from executives or suppliers, especially payment requests. Call back on a known number.
- Label your own AI imagery. If your website uses AI-generated product or blog images, say so. It builds trust and matches where regulation is heading.
- Keep originals of your real product photos and brand assets, so you can prove what's yours.
Building trustworthy websites
At NextGen Fusion, our web development company in Lucknow, we help brands handle AI-generated content responsibly, from image policies to clearly labelled AI features on their websites.
