I'd totally missed that SpaceXAI/SpaceXSI/xAI (or whatever they're called now) had turned on text watermarking by default through the Grok API starting on November 7, there hadn't been much chatter about it, no formal announcement of sorts. I wasn't even sure they were going to follow OpenAI/Anthropic et al, though I guess they had little choice due to EU regulations.

However, it now seems like they have turned it OFF by default, some time in the past few days as per the note at the top of the documentation. So admins now have to go into the console and turn it on.

Not sure what difference any of this makes tbh, but kinda interesting the different approaches the various AI labs are taking with API text watermarking. Anthropic is on by default with no option to turn off (I think?), something similar for Google/Gemini, OpenAI's is optional and off by default. Seems Grok had gone for a sort of hybrid, "on by default but with the ability to turn off". And now it's off by default.