Google expands SynthID: invisible labeling of AI content is becoming a standard

At the Google I/O conference, the company announced a major expansion of SynthID, its technology for labeling AI content. It is an attempt to turn back a wave of fake images and videos that have already begun to shape how people perceive reality on social media.

How the invisible labeling works

SynthID is an invisible watermark embedded in the pixels of an image or in an audio track when AI models create the content. The human eye cannot see it, but a verification system can detect the metadata. It works like a signature that a system can read but that does not degrade the content.

In parallel, the C2PA standard (Coalition for Content Provenance and Authenticity) is being developed. It is an industry consortium that registers in advance where and when content was created. The two technologies complement each other: SynthID is the internal watermark, C2PA is the public provenance history.

• Invisible labeling is added automatically during generation in Gemini, Imagen and other Google AI products
• Verification works directly in the browser via Chrome and on mobile via Android
• The C2PA standard is supported by YouTube, OpenAI, Meta* and other major players

The scale of the problem

Until now, the system operated in a semi-secret mode. Google's expansion means millions of users will now get a verification tool in the places they already use.

The viral story of Pope Francis in a designer coat showed how fast AI-faked images spread. It racked up millions of views within hours before people realized it was fake. Had SynthID been in place, checking would have taken a second.

This is not just entertainment. Disinformation campaigns use AI fakes to manipulate opinion ahead of elections, eroding trust in the media. The first elections under pressure from synthetic content are happening right now.

"We must find a balance between protecting against disinformation and preserving creative freedom," is the gist of

Google's position.

Where it will be available

Chrome will add a built-in verification tool for images. YouTube will label videos containing AI-generated content. Android will show an authenticity badge in the gallery. Google Lens (image search) will be able to identify labeled content automatically.

But the system has gaps. SynthID works for content created by Google AI. Models from OpenAI, Meta, Midjourney need their own labeling. And if any system does not build in verification, fakes will still remain.

What this means

We are entering a critical period in which truth and falsehood online will become not a matter of trust but a matter of tools. SynthID and C2PA are first steps toward a world where the authenticity of media is verified like a payment.

*Meta has been designated an extremist organization and is banned in Russia.