article --summary
The summary
This piece reports that SpaceXSI's Grok API documentation now lists text watermarking as disabled by default, reversing an earlier notice that it would switch on automatically in early November. Developers can turn it on per model through team admin settings, though the company has offered no public explanation for the shift.
The article places the move in context of the EU AI Act's machine-readable marking rules for synthetic text, and compares how Anthropic, OpenAI and Google have handled watermarking. It also notes that SpaceXSI's documentation confirms it uses Google's SynthID Text technology.
The practical upshot is that watermarking is controlled per team rather than per request, so developers serving different regions would need separate team setups to apply different policies.
Editorial note: This is a summary of an article first published by The New Stack. The full piece is on their site.