Grok’s API text gets an invisible watermark by default on 7 November
SpaceXSI’s developer page says seven models, including grok-4.7, will shape token choices so the company can later score a passage as watermarked, not watermarked, or uncertain. Callers send no new parameter. Team admins can opt out per model, after confirming a legal exception, and some models can be locked on.
Feature2 minEasy
Maya Chen2 min read
- Grok
- xAI
- API
- EU AI Act

Photo: Runtime WireSource page
SpaceXSI’s developer documentation says that from 7 November 2026, text from seven Grok models will carry an invisible watermark by default. The page lists grok-4.7, grok-4.6, grok-4.5, grok-4.3, and three grok-4.20 variants: reasoning, non-reasoning, and multi-agent. Ryan Merket wrote the page up for Runtime Wire at 12:43 p.m. Central time on Saturday, and updated the story at 3:08 p.m. The documentation does not print a clock for when the page itself changed. It does print the date the mark turns on.
The mark is not a label a reader can see, and it is not a string hidden in the text. The page says it is applied while the model generates: it slightly shapes which tokens the model picks, and it adds no characters, markup, or hidden text. A token is one of the chunks the model chooses from as it writes. Nudging that choice leaves a statistical pattern. Detection, the page says, is a score. SpaceXSI reports one of three results: watermarked, not watermarked, or uncertain. Longer text gives a stronger signal. Access to that detection is by application. It is not a public endpoint.
What a caller actually receives
The watermark applies to text from the Responses API and the Chat Completions API, on the global endpoint, https://api.x.ai, and on any regional endpoint. It covers ordinary replies, structured output, and the arguments of tool calls, which is the part of a response an agent uses to invoke a function rather than to show a paragraph to a person. A caller does not send a new parameter. The response format stays the same except for a header, x-watermark. A value of true means that output was generated with the mark. Any other value, or no header, means it was not. The FAQ on the same page says watermarking does not change response quality or speed.
Team administrators can turn the mark off per model in the SpaceXSI console. The setting applies to every API key on the team. The page says a change reaches new requests within five minutes, and that opting out asks the admin to confirm that an exception under applicable law applies and that the team is solely responsible for any marking or disclosure requirement. Some models may be locked on. Those appear in the dialog without a checkbox. Every opt-in and opt-out is written to the team’s audit log. The reason the page gives for the mark is compliance with the EU AI Act. It does not say whether the consumer Grok apps are in the change. The page is filed under API usage.


