xAI and Grok watermarks: the announced API rollout

xAI has announced text watermarking for supported Grok API models from 7 November 2026. Overwride has opted out because we use our own ledger for disclosure.

What xAI has announced

The API documentation schedules default watermarking for models including grok-4.7. It covers Responses and Chat Completions, including structured output and tool arguments. The signal changes token selection without adding characters. xAI API docs: output watermarking.

Read this as an announced rollout, rather than evidence that a particular response generated before that date was marked. The claim “Grok never watermarks text” is no longer a sound basis for choosing a writing workflow.

How to check the documented setting

The documented x-watermark: true response header identifies a marked API output. Team administrators control eligible models through team settings; some models may be locked. Opting out requires confirming an applicable legal exception and responsibility for disclosure. Detector access is by application. xAI API docs: output watermarking.

For an integration, preserve response context when you need to understand how a result was produced. A setting for future generation and a detector examining existing text serve different purposes. Changing a generation setting does not inspect a draft you pasted into a tool.

For your own integration, check your team’s settings and response headers for the models you use.

Text, images, and video are different cases

The Grok consumer FAQ separately documents visible watermarks on generated images and video, with no removal setting. That media policy does not explain the API’s text signal. Grok FAQ: image and video watermarks.

Overwride works with text. It does not remove logos from pictures, edit video frames, or strip media credentials. A text guide should not imply that one operation handles every kind of “Grok watermark.”

Why Overwride uses its own ledger

Overwride has opted out of Grok’s text watermarking because we use our own ledger. The ledger keeps the submitted draft and returned rewrite together in a shareable record. Readers can follow the disclosure link and compare the versions without a provider’s detection key.

The opt-out controls watermarking during Grok generation. It does not test whether a submitted draft contains a watermark from another provider. The ledger documents the rewrite step and gives readers the actual text to inspect.

A practical publishing workflow

Keep your starting draft, review the rewrite for changed meaning, and retain the version you publish. Describe the assistance you used in ordinary language. If you share an Overwride record, remember that readers can see its original draft too; avoid including private notes you do not intend to disclose.

The Overwride versus xAI watermarking comparison explains how generation signals and reader-facing records fit into that process. Developers can also read the rewrite API guide.