Google Removes Visible AI Watermarks, Relies on Invisible Provenance Not Yet Widely Detected

On August 14, 2026, Google let users remove the visible Gemini sparkle from AI-generated images and video, keeping invisible SynthID and C2PA metadata intact. The toggle is excluded in the EU and South Korea, where visible watermarks are legally required. No SynthID detector runs at most consumption points — social feeds, platforms, messaging apps. Reuters found cropping can defeat SynthID in July 2026 testing.