According to The Verge, Google rolled out a "Media watermark" toggle that strips the visible mark from AI-generated content. The files retain hidden SynthID signals and C2PA metadata in the background, but the human-readable badge disappears on first glance.
Visible labels work for people, not forensic tools. Most users never inspect metadata, and most platforms bury it anyway. This change makes AI content easier to pass off as human-made. That burden lands on viewers, publishers, teachers, and anyone trying to verify what they're looking at in real time.
Google didn't remove watermarking. It converted it from human-readable to machine-readable, which helps audits after the fact but fails at the exact moment a fake image gets screenshotted, reposted, and stripped of context across platforms.
Google is betting that provenance should be invisible until someone checks for it. That serves creators who hate corner badges, but it also serves spammers, engagement farmers, and anyone who wants AI content to blend in on first sight. If the company wants trust points, hidden markers aren't enough. Put the disclosure where actual humans can see it—inside the products where actual humans share it. Don't bury it in metadata and leave it for Gemini Search to explain later.
Filed to the Technology desk · Aug 14