Free guide

Grok API Watermark: What Changed on Nov 7 and How to Check It

Since November 7, 2026, text from seven Grok models on the API carries an invisible watermark by default. What it is, how it works, how to check your own requests, and what turning it off really means.

By Beau Johnson·
The Short this guide goes with: Grok API Watermark: What Changed on Nov 7 and How to Check It

Quick answer

Starting November 7, 2026, SpaceXSI (the company behind Grok) watermarks the text output of seven Grok models on its API by default, to comply with the EU AI Act. The watermark is invisible: it slightly shapes which tokens the model picks and adds no characters. You can see it on your own requests in the x-watermark response header, and only team admins can opt out, per model, by confirming a legal exception applies.

The short version

  • Seven Grok API models watermark their text by default since November 7, 2026: grok-4.7, grok-4.6, grok-4.5, grok-4.3 and the three grok-4.20 models.
  • Nothing is added to the text. The watermark lives in which words the model picks, and detection is statistical, so longer text gives a stronger signal.
  • Check any request with the x-watermark header. Turning it off is a per-model team setting that asks an admin to confirm a legal exception, and it lands in the audit log.

How it actually works

It lives in the word choices

SpaceXSI's docs say the watermark is applied while the model generates text: it slightly shapes which tokens the model picks. No characters, markup or hidden text get added, there's no parameter to send, and the response format stays the same apart from one new header.

Think of a cipher card with holes in it

Picture a card with holes punched in it. Lay it over a page and count how many words land in the holes. On an ordinary page you'd get about what chance gives you; way more than that and the page gets flagged. Detection works in a similar statistical spirit: SpaceXSI scores a piece of text against its watermark and reports watermarked, not watermarked, or uncertain. Access to that detection is granted by application, and longer text gives a stronger signal.

What it covers

Text output from text generation requests on every SpaceXSI API endpoint, including api.x.ai and regional endpoints. That includes the Responses API and Chat Completions API, ordinary text replies, structured outputs and tool call arguments. SpaceXSI says it does this to comply with the EU AI Act, and its docs say it doesn't change response quality or speed.

Turning it off is a legal call

Team admins can opt out per model in the SpaceXSI Console, and the setting applies to every API key on the team. Opting out asks you to confirm that an exception under applicable law applies to your use and that you're solely responsible for any marking or disclosure. Some models may be locked on, and every opt-in and opt-out is recorded in the team's audit log.

How to set it up

  1. 1.Check one of your own requests

    Send any request and read the response headers. x-watermark: true means that output was watermarked. Any other value, or no header, means it wasn't. When we ran this on November 8 with grok-4.3 (a short customer-support reply), the header came back true.

    curl -i https://api.x.ai/v1/responses \
      -H "Authorization: Bearer $XAI_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{"model": "grok-4.7", "input": "Write a haiku about the ocean."}'
    # look for: x-watermark: true
  2. 2.Log the header in your app

    If you use the OpenAI Python SDK pointed at SpaceXSI, read the raw response to get the header, then store it next to each generated message so you know what went out marked.

    raw = client.responses.with_raw_response.create(
        model="grok-4.7",
        input="Write a haiku about the ocean.",
    )
    print(raw.headers.get("x-watermark"))
    response = raw.parse()
  3. 3.See your team's setting

    Team admins: open Team settings in the SpaceXSI Console and find the Model output watermarking row. It shows On (all models), Off (all models) or Custom. The settings icon opens the per-model list. Only change it after your legal or compliance owner has signed off, because opting out asks you to confirm a legal exception applies.

  4. 4.Check the audit log

    Every opt-in and opt-out is recorded on the Audit Log page in the SpaceXSI Console. Search by event ID, description or user and filter by date to see who changed what.

  5. 5.Need to verify text? Apply for detection

    Detection isn't open to everyone. SpaceXSI grants access to its watermark detection API by application, through the form linked from the watermarking docs.

The fine print

  • This is the Grok API. The docs cover API text output from the seven listed models, not the Grok app or X.
  • Opting out asks a team admin to confirm a legal exception applies and that you're solely responsible for any marking or disclosure. It isn't a developer preference toggle.
  • Some models may be locked on for your team and can't be turned off from the Console.
  • Changes apply to new requests within 5 minutes and cover every API key on the team.
  • Detection is statistical (watermarked, not watermarked, or uncertain). Short text gives a weaker signal.
  • Grok isn't the only one. Anthropic watermarks Claude's text for the EU AI Act too, with its own key and its own detector.

FAQ

Is Grok API text watermarked?

Yes, since November 7, 2026, for grok-4.7, grok-4.6, grok-4.5, grok-4.3, grok-4.20-reasoning, grok-4.20-non-reasoning and grok-4.20-multi-agent. It's on by default on every SpaceXSI API endpoint.

Can you see the watermark in the text?

No. It adds no characters, markup or hidden text. It slightly shapes which words the model picks while it writes, so the text reads the same.

How do I know if a response was watermarked?

Read the x-watermark header on the Responses API or Chat Completions API response. true means it was watermarked; any other value or no header means it wasn't.

Can I turn the Grok watermark off?

Team admins can opt out per model in the SpaceXSI Console, unless a model is locked on. Opting out asks you to confirm a legal exception applies to your use and that you're solely responsible for any marking or disclosure. It's recorded in the audit log.

Does the watermark change quality, speed or my code?

SpaceXSI says it doesn't change response quality or speed, and there's no parameter to send. The only change to the response is the x-watermark header.

Sources

Facts checked 2026-11-08 against SpaceXSI's Output Watermarking docs (last updated Oct 31, 2026) and our own grok-4.3 API call on Nov 8, which returned x-watermark: true.

Want help building with tools like this?

Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.

Join Shipping Skool

More free guides