November 1, 2026

Grok API Output Watermarking: What Changes on 7 November 2026

Written by Christos Papadimitriou November 1, 2026

Short answer: From 7 November 2026, text generated through the Grok API by supported models (grok-4.3 to grok-4.7 and the grok-4.20 family) carries an invisible watermark by default. You do not need to change any code, and you can check each response through the x-watermark header. Team admins can opt out per model, but opting out asks you to confirm that a legal exception applies and that your team is solely responsible for any marking and disclosure.

If you run AI agents or content workflows on the Grok API, especially in the EU and Greece, check this before 7 November. Opting out is a legal decision, not a technical one.

All facts about the watermark come from the public developer documentation of SpaceXSI, the company named on the Output Watermarking page, as read on 31 October 2026.

What changes on 7 November 2026

According to the documentation, “Starting November 7, 2026, responses from these models include a watermark. It’s on by default, and you can opt out per model.” The models listed are:

grok-4.7
grok-4.6
grok-4.5
grok-4.3
grok-4.20-reasoning
grok-4.20-non-reasoning
grok-4.20-multi-agent

SpaceXSI states that it watermarks text model output “to comply with the EU AI Act”, and that the watermark is enabled on every API endpoint, including the global endpoint https://api.x.ai and any regional endpoint. The page does not list which regional endpoints exist, so do not read this as a statement about an EU endpoint.

What gets watermarked in your agents and workflows

The watermark applies to text output from every text generation API, including the Responses API and the Chat Completions API. It covers the generated text you receive back:

ordinary text responses;
structured outputs (for example, JSON your workflow parses);
tool call arguments (what your agent passes to its tools).

That last point matters for agents: a customer email, a CRM field or a support reply written by Grok is in scope. The page covers text output only.

Does it change quality, speed or my code?

No code change is needed: there is no request parameter, and the response format is unchanged apart from the x-watermark header. On quality and speed, SpaceXSI’s answer is: “No. Watermarking doesn’t change response quality or speed.” That is the vendor’s claim; if output quality is critical for you, run your own evaluation set before and after 7 November.

How the watermark works

The watermark is not visible. Per the documentation, it is applied while the model generates text: it “slightly shapes which tokens the model picks, so it adds no characters, markup, or hidden text to the response.”

Detection is statistical. SpaceXSI scores a piece of text as watermarked, not watermarked or uncertain, and longer text gives a stronger signal. Access to the detection API is granted by application.

The page does not say how the watermark behaves after editing or translation, so treat it as one signal, not proof.

How to check whether a response was watermarked

Read the x-watermark response header. x-watermark: true means that request’s output was generated with the watermark. Any other value, or no header at all, means it was not.

A manual test, from the documentation:

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."}'

In production, log the header with each call. With the OpenAI Python SDK (as in the documentation):

raw = client.responses.with_raw_response.create(model="grok-4.7", input=prompt)
response = raw.parse()
watermarked = raw.headers.get("x-watermark") == "true"
# store `watermarked` with the request ID, model, date and where the text went

Opting out: why it shifts legal responsibility to you

Team admins can turn watermarking off for some or all models in the SpaceXSI Console (Team settings, “Model output watermarking”). Key facts from the documentation:

The setting is per model and applies to every API key on the team. You cannot opt out for one project’s key only.
Changes apply to new requests within 5 minutes.
Opting out “asks you to confirm that an exception under applicable law applies to your use, and that you are solely responsible for any marking, disclosure, or other legal requirement.”
Some models may be locked on for your team and cannot be turned off from the Console.
Every opt-in and opt-out is recorded in the team’s audit log.

In plain terms: opting out does not remove a legal duty. It moves it from the provider’s built-in marking to your own team. If you are weighing who runs your agents in the first place, see our comparison of DIY agents on Cursor or Grok Bot vs OS47.

When you should NOT opt out

You publish Grok-generated text to the public (blog posts, newsletters, social posts, customer-facing replies) and have no other marking or disclosure process.
You cannot name the specific legal exception that applies to your use, with your lawyer’s sign-off.
Several teams or clients share one API team, so one admin’s choice would change marking for everyone’s keys.
You have no owner for the decision and no record of why it was made.

How this connects to the EU AI Act (Article 50)

Article 50 of the EU AI Act, Regulation (EU) 2024/1689, sets transparency obligations for providers and deployers of certain AI systems. Under Article 50(2), providers of AI systems that generate synthetic text and other content must ensure the outputs are marked in a machine-readable format and detectable as artificially generated. Article 50(4) asks deployers that publish AI-generated text to inform the public on matters of public interest to disclose it, unless the text has gone through human review or editorial control and someone holds editorial responsibility.

According to the European Commission’s AI Act Service Desk, transparency obligations apply and are enforceable from 2 August 2026. A provider-side watermark does not replace your own disclosure duties as a deployer, so the decision belongs in your AI governance framework, next to your other guardrails. Our post on who owns the AI agent decision covers the deployer role.

This is not legal advice. Check your own case with a lawyer.

A practical checklist before 7 November

Inventory where Grok output goes. Every agent and workflow that calls the Grok API, its model, and where the text ends up.
Log the x-watermark header with model, date and request ID for every call.
Decide the per-model policy. Keep watermarking on by default; any opt-out needs a named legal exception and an owner.
Document the decision. Record who decided, why, and when. Cross-check it against the Console audit log.
Align with your own disclosure labels. The watermark is machine-readable; people still need the labels your policy requires. Our AI agent security guide covers the wider controls.
Re-check after 7 November that the header appears on the models you expect, and that output quality still passes your tests.

Still deciding whether you need agents at all? Start with our guide for Greek CEOs. For how we handle approvals, logs and data, see Trust & safety.

Want a person to sign off before AI-written text goes out? OS47 is theagency47’s client workspace for AI agents: a private web app where your team sees what your agents know and do, approves every reply and knowledge change before it is sent, checks each agent against test questions with known answers, and can stop every agent at once. OS47 records who decided and when, and it is included at no separate price with every theagency47 product. See how OS47 approvals and agent controls work.

FAQ

Does the Grok API watermark text output?

Yes. From 7 November 2026, text output from supported Grok models carries an invisible watermark by default, according to SpaceXSI documentation. It covers the Responses and Chat Completions APIs, including structured outputs and tool call arguments.

Which Grok models are watermarked?

The documentation lists 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. Watermarking is on by default for these models from 7 November 2026, and team admins can opt out per model unless a model is locked on.

Do I need to change my code for Grok API watermarking?

No. There is no request parameter and the response format is unchanged, apart from a new x-watermark response header. Existing integrations keep working as before.

How do I check whether a Grok API response was watermarked?

Read the x-watermark response header. A value of true means the output was generated with the watermark; any other value, or no header, means it was not.

Can I see the watermark in the text?

No. The watermark is applied during generation by slightly shaping which tokens the model picks, so it adds no characters, markup or hidden text. Detection is statistical, works better on longer text, and the detection API is available by application.

How do I opt out of Grok API watermarking, and should I?

Team admins can opt out per model in the SpaceXSI Console (Team settings, "Model output watermarking"); the change applies to every API key on the team, some models may be locked on, and every change is recorded in the team audit log. Usually you should not: opting out asks you to confirm that a legal exception applies and that your team is solely responsible for any marking and disclosure. Under EU AI Act Article 50 that responsibility is real. This is not legal advice.

theagency47 is not affiliated with, endorsed by, or a partner of SpaceXSI. Grok is a trademark of its owner. Facts in this article come from SpaceXSI’s public developer documentation as of 31 October 2026 and may change.

Sources: Output Watermarking, docs.x.ai · Regulation (EU) 2024/1689, EUR-Lex · When does enforcement start?, AI Act Service Desk · Transparency obligations under Article 50, European Commission (all accessed 31/10/2026)