Starting today, 7 November 2026, text generated by seven Grok models through the xAI API carries an invisible watermark by default. SpaceXSI — the SpaceX unit that now builds Grok, formerly xAI — documents the change on its "Output Watermarking" developer page, last updated 31 October, and gives a one-line reason: it watermarks text "to comply with the EU AI Act."

The watermark applies to grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three grok-4.20 variants (reasoning, non-reasoning and multi-agent), across the Responses API and the Chat Completions API, on every endpoint including the global api.x.ai host and regional endpoints. It covers ordinary text as well as structured outputs and tool-call arguments. The mark is a statistical pattern in token selection, not hidden characters or markup — readers see nothing different, and the company says it does not change response quality or speed, a claim no independent party has yet measured. The only visible change is a new response header: x-watermark: true means that request's output was watermarked.

Opting out is possible but deliberately frictionful. Team admins can switch watermarking off per model in the SpaceXSI Console, but only after confirming that a legal exception covers their use and that they alone are responsible for any marking or disclosure — and every change is logged. Some models may even be locked on for a team, with no checkbox to turn them off. The timing is no accident: Article 50 of the EU AI Act requires providers of generative AI systems to mark synthetic content in a machine-readable, detectable way, with a grace period ending 2 December 2026 for systems already on the market — 7 November falls 25 days before that deadline.

Grok is the second major lab to flip the switch this week: reports say Anthropic extended the same kind of mark to four more Claude models two days earlier, and OpenAI says ChatGPT is getting one in the EU only. Two limits matter for readers. First, the change covers the developer API only — the docs say nothing about the Grok app, grok.com or Grok on X. Second, there is no public detector: like Anthropic, xAI limits detection to approved organisations such as regulators, researchers, media and fact-checkers, and none of the three companies has published real-world error rates for its text mark. The watermark is there, but for now only the labs can see it.

Sources