The biggest real-world test of AI content labeling is about to begin. SynthID, Google's invisible watermarking technology, and C2PA Content Credentials, an open industry standard for tracking a file's origin and edit history, are rolling out their largest expansions yet - and this is where we find out if either approach actually holds up outside a lab.
The two systems work differently. SynthID embeds an imperceptible signal directly into the pixels of an image, the frames of a video, or the waveform of an audio file. Even after compression, cropping, or re-encoding, the watermark is supposed to survive and remain detectable. C2PA Content Credentials attach cryptographically signed metadata to a file - essentially a tamper-evident certificate that records where the content came from, what tools created or modified it, and when. Both can, in theory, tell a viewer whether what they're looking at was made or altered by AI.
The problem is that neither works unless the tools people actually use adopt them, and neither is useful to readers unless they have software that can read the tags. A watermark no one checks is theater.
Where the Standard Actually Stands
C2PA has significant industry backing. Adobe built Content Credentials into Photoshop and Adobe Firefly. Microsoft, Sony, the BBC, and Truepic are all signatories to the spec. The standard is now on version 2.1, and the Content Authenticity Initiative - the Adobe-led industry group pushing C2PA adoption - claims over 3,000 members.
SynthID, developed by Google DeepMind, has expanded beyond images into audio, video, and text. Google has integrated it into Gemini and several of its own AI products. The current expansion moves both systems into broader platform deployment across more content types and more users.
The Weak Links
There are real problems with both approaches that wider adoption won't automatically fix.
C2PA metadata can be stripped. Save a C2PA-credentialed image as a new JPEG without the credentials attached, and the certificate disappears. The spec accounts for this - a missing credential is supposed to be a red flag in itself - but that only works if viewers know to look for the absence.
SynthID watermarks are more durable but not indestructible. Academic researchers have shown that adversarial image processing can degrade watermarks, and detection requires access to the same model layer that created the watermark in the first place.
There's also a fragmentation problem. SynthID is a proprietary Google technology. C2PA is an open standard. They don't interoperate, and a platform choosing one doesn't get the benefits of the other.
The real test isn't technical - it's whether the major social platforms enforce credential checking at upload time and whether browsers or mobile apps surface that information to ordinary users. Without that last-mile adoption, both technologies remain metadata that only journalists and researchers will ever see. This expansion is large enough that adoption rates will become measurable, and the gaps will be harder to dismiss.