Text generated through Grok's developer API began carrying an invisible watermark on 7 November 2026, after the company behind the chatbot switched the marking on by default for seven of its models. The firm, formerly xAI and now operating as SpaceXSI, says it watermarks text output to comply with the EU AI Act, according to its developer documentation on output watermarking.

The change covers grok-4.7, grok-4.6, grok-4.5 and grok-4.3, plus three versions of grok-4.20 for reasoning, non-reasoning and multi-agent work. It applies to every text generation API, including the Responses API and the Chat Completions API, on the global endpoint and regional endpoints alike. All output types are covered, with no change needed from developers.

The watermark cannot be seen. SpaceXSI says it slightly shapes which tokens the model picks while generating, so it adds no characters, markup or hidden text to a reply. The only visible trace is a new response header: x-watermark: true means that request's output carries the mark. Detection is statistical.

Team administrators can switch the watermark off, model by model, in the SpaceXSI Console, with changes taking effect within five minutes. But opting out comes with conditions: the admin must confirm that an exception under applicable law covers the use and that the team is solely responsible for any marking or disclosure required. Every change is logged, and some models may be locked on.

The timing sits ahead of the EU AI Act's transparency deadline: systems already on the market before 2 August 2026 have until 2 December to mark synthetic content in a machine-readable way. Anthropic watermarks text from its recent models with no way to turn it off, while OpenAI leaves its API watermark off by default. The Grok change covers the API only, not the Grok apps or Grok on X.