Grok to receive invisible text watermark from 7 November
xAI says supported Grok models will make their text output identifiable as AI-generated.
Listen to this article
There is no audio version yet. Request one and an AI voice will read the article aloud.
Read by an AI voice.
.webp){kind=link}
From 7 November, xAI will add an invisible watermark to text generated by supported Grok models. The feature is enabled by default, but administrators can disable it per model, according to the documentation.
According to the technical documentation, the watermark is embedded in the text output itself and requires no changes to users’ requests. xAI says the signal may later help identify text generated by a model.
The measure applies to several Grok models, including Grok 4.5, 4.6 and 4.7. Output from the Responses API, Chat Completions API, structured responses and tool calls is also covered by the announced functionality, according to xAI.
xAI links the introduction to compliance with the European AI Act. At the same time, the documentation states that team administrators can disable the feature per model. It is therefore not a given that every piece of Grok output will always carry a visible or persistent provenance signal.
A text watermark is different from a conventional detector that looks only at style or word choice after the event. The aim is for a system to recognise a statistical or technical pattern in the generated text. Its practical reliability will depend on the final implementation, subsequent editing and whether the text is combined with human editing.
For newsrooms, schools and businesses, the introduction may be relevant when they want to establish how digital content was produced. The watermark does not automatically prove who wrote a text or that its content is accurate. At most, it says something about the provenance of the output when the signal remains recognisable.
One story, several perspectives
What is established
- xAI is announcing an invisible watermark for the text output of supported models.
- The feature is enabled by default, but can be disabled per model.
- xAI links the measure to the European AI Act.
Left
Arguments Emphasises transparency about AI use, protection of creators and public oversight of synthetic content.
Values Accountability, consumer protection and limits on the power of technology companies.
Consequences Fears that voluntary deactivation and limited oversight will be insufficient for education, journalism and elections.
Centre
Arguments Emphasises technical standards that enable provenance information, while allowing room for exceptions and evaluation.
Values Practicality, privacy, innovation and the enforceability of regulation.
Consequences Wants to prevent watermarks from being treated as conclusive evidence when text has been edited or passed through multiple systems.
Right
Arguments Emphasises freedom of choice for businesses and users and warns against mandatory technology that could restrict innovation or freedom of expression.
Values Individual freedom, market forces and restrained regulation.
Consequences Fears additional compliance costs and a system in which platform companies indirectly determine which content is considered authentic.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The technical claims were checked directly against xAI’s official documentation. The text avoids the unsubstantiated conclusion that watermarks are always reliable or indelible.
- confirmed Grok text will receive a watermark by default from 7 November 2026. — xAI explicitly states the date and the default setting. source
- confirmed The feature applies to several Grok models and API formats. — The documentation names the models and API output. source
- confirmed Administrators can disable the watermark per model. — xAI mentions an opt-out per model. source
- confirmed The watermark does not prove who wrote a text or that its content is accurate. — This is a limitation of what provenance technology can logically demonstrate; xAI does not claim authorship or truth verification. source
Editor's note
The date and technical operation come from xAI’s own documentation. Independent information on detection accuracy and actual use in consumer products is still lacking.Sources
- Output Watermarking — xAI
- xAI Adds Invisible Text Watermarking To Grok API Models — Crypto Briefing
More on this in Dutch media
- de Volkskrant — „grok xai”
- NOS — „grok xai”
- Het Parool — „grok xai”