Grok now watermarks its text invisibly: what changes for your business from November 7
Direct answer
Starting November 7, 2026, text generated by seven Grok models through xAI's API carries an invisible watermark switched on by default. The company, which appears in its documentation as SpaceXSI, says it does so to comply with the EU AI Act, which requires AI-generated content to be marked in a machine-readable way. The watermark does not change what the text says or its format: it is verified by reading the x-watermark header of the response. If your product calls the Grok API and you did not touch the setting, your responses already carry the mark, without you writing a line of code.
What exactly changed
- Seven models covered: grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three grok-4.20 variants (reasoning, non-reasoning and multi-agent). The mark applies on every API endpoint, including the global api.x.ai and regional ones.
- Two APIs, all the text: the mark covers the Responses API and the Chat Completions API, and reaches generated text, structured outputs and the tool-call arguments Grok fills in.
- On by default, no code change: there is no parameter to send. The response arrives the same, with one new header: x-watermark: true means that response carries the mark.
- Turning it off is an admin decision: only team admins can switch it off, model by model, from the console, and they must confirm that a legal exception covers their use.
- API only, for now: the documentation covers the developer API. For the Grok app, the website and Grok inside X there is no announced date, and neither for Grok Imagine images and videos, which carry a different mark.
Why this matters even if you do not use Grok
- The EU deadline is December 2, 2026: providers whose systems were on the market before August 2 must meet the Article 50 marking obligation by that date. Grok arrived 25 days before the deadline; other providers have to get there too.
- The mark is detectable, but the detector is not public: xAI offers a detection API that answers whether a text is watermarked, not watermarked or uncertain, but access is by application. Today an ordinary reader cannot check on their own whether a Grok text carries the mark.
- OpenAI marked its text first: in October, OpenAI started invisibly watermarking ChatGPT text, also for the AI Act. Marking AI-generated content is becoming the industry standard, not one vendor's exception.
- Check your integrations: if your agent generates text that you then publish, send to customers or use as a document, that text may carry a mark. Ask your vendor what it marks, where and who can verify it, before a customer or a regulator asks first.
Note: this article summarizes xAI's developer documentation ("Output Watermarking", last updated October 31, 2026) and independent coverage of that change. The December 2, 2026 deadline applies to systems on the market before August 2, under Article 111 of the AI Act as amended by the Digital Omnibus package. The promise that the watermark changes neither quality nor speed comes from xAI and has no published independent testing.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- OpenAI started invisibly watermarking ChatGPT text: what it means for your business
- The EU already fines for AI: up to 35M euros — what to know if you sell to Europe
- How to use AI in your content without losing your voice
Frequently asked questions
What is Grok's watermark?
An invisible signal inside the text Grok models generate through the API. It adds no characters and does not change the content: it slightly shapes which tokens the model picks, and it is verified through the x-watermark header of the response.
Which models does it cover from November 7?
Seven: grok-4.7, grok-4.6, grok-4.5, grok-4.3 and the three grok-4.20 variants (reasoning, non-reasoning and multi-agent), on every API endpoint.
Can I turn the watermark off?
Only a team admin, model by model, from the SpaceXSI console, confirming that a legal exception covers that use. Turning it off is not an option for an individual user.
Why is xAI watermarking its text?
To comply with the EU AI Act, whose Article 50 requires AI-generated content to be marked in a machine-readable way and detectable as artificially generated. The compliance deadline for systems already on the market is December 2, 2026.
