Grok API Now Watermarks Every Text Output by Default
In short
- Since November 7, the xAI Grok API adds an invisible watermark to the text of seven models by default.
- You change no code. A response header called x-watermark tells you if a reply was marked.
- The money angle: opting out needs a team admin and a legal exception, so a wrong guess costs you a compliance problem, not a few euros.
If you build anything on the Grok API, your outputs now carry a hidden signature. You did not ask for it and you cannot see it. Here is what it means for you.
What happened
According to xAI’s developer docs, watermarking is on by default for text output starting November 7, 2026. It covers seven models: 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 works on both the Responses API and Chat Completions. It also covers structured outputs and tool call arguments. The docs say the reason is EU AI Act compliance.
The mark adds no characters and no markup. The docs say it does not change quality or speed. Detection is statistical: a checker answers watermarked, not watermarked, or uncertain. Longer text gives a stronger signal. Access to the detection API is by application.
How to use it
First, check if you are affected. Read the x-watermark response header. If it says true, the reply was marked. Any other value, or no header, means it was not.
Second, decide if it matters. Most small businesses will not care. A product description or a support reply keeps working exactly as before. It matters if you sell text as a human product, for example ghostwriting, or if your clients’ contracts say the text must be unmarked.
Third, if you need out, only a team admin can do it. Go to Team settings in the xAI console, switch it off per model, and confirm that a legal exception applies. You accept sole responsibility for any related requirements. The setting hits every API key on the team and takes effect within 5 minutes. Changes are logged. Some models are locked on and cannot be turned off.
Do the math: the docs list no extra price. Building your own detection would be real money. A freelance developer at €50 an hour for 20 hours is €1,000. The header costs you €0. That is an assumed freelancer rate, not an xAI figure.
What it means
Europe’s rules for marking AI text are starting to show up inside APIs. xAI did it as a default, and the exit needs a legal reason. Expect other providers to follow with their own versions.
One honest limit: the docs explain how to check a mark, not how robust it is. Nobody published how well it survives heavy editing. Do not build a business process that depends on it surviving.
Ivan’s Take
For me this is a non-event for normal business use. I would add the x-watermark check to my logs, so I know what leaves my system, and move on. I would only fight it if a client contract needs unmarked text, and then I would pick a provider without this default instead of hunting for a legal exception.
Source: xAI
If it doesn’t save you time or money, you won’t see it here
- Facts, not hypeEvery price and number is checked against the official source
- Always the real priceYou see what it costs before you click
- I say when it’s weakEven when it’s OpenAI or Google
