Grok Text Will Get Invisible Watermarks to Follow EU AI Rules

SpaceXAI will start adding invisible watermarks to text generated by its Grok models from November 7. The change covers APIs in all regions and is meant to follow the EU AI Act. A detector can read the signal and tell if a piece of text comes from a watermarked model.

The watermarked models include grok-4.7, grok-4.6, grok-4.5, grok-4.3, and several grok-4.20 versions. Team admins can still turn off the watermark for specific models, keeping some flexibility in how they use them.

Instead of hiding characters or metadata, this watermark is built into how the model picks tokens while writing. It cannot be seen by the eye.

AI governance watcher Andrew Curran says ChatGPT in North America is now the only model that does not add invisible watermarks to text, as AI content transparency rules keep speeding up.