Text generated through xAI’s Grok API will carry an invisible watermark by default from 7 November. The Grok API watermark covers seven current models, needs no change to developers’ requests, and is there, the company says, “to comply with the EU AI Act”. Teams can switch it off, but only by declaring that the legal duty is theirs.
The change is set out in a new Output Watermarking page in xAI’s developer documentation, which now carries the SpaceXSI name and was last updated on 31 October. “Starting November 7, 2026, responses from these models include a watermark. It’s on by default, and you can opt out per model,” it says.
Which Grok models and APIs are covered
The list is grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three versions of grok-4.20: reasoning, non-reasoning and multi-agent. The mark applies through both the Responses API and the Chat Completions API, and covers ordinary text answers, structured outputs and the arguments a model writes when it calls a tool. “The watermark is enabled on every SpaceXSI API endpoint, including the global endpoint, https://api.x.ai, and any regional endpoint,” the documentation says.
Developers can tell whether a reply was marked from a response header, x-watermark: true. The page does not mention the consumer Grok apps. TokenPost reported that the change applies to the API and does not affect consumer-facing Grok products.
How the invisible mark works, and who can check it
Nothing is added to the text. According to the documentation, the watermark “slightly shapes which tokens the model picks, so it adds no characters, markup, or hidden text”. A token is the small chunk of text a model produces at each step, so the mark lives in a statistical pattern in word choice, not in anything a reader could see or delete.
Detection is statistical as well. A check returns one of three answers, watermarked, not watermarked or uncertain, and “longer text gives a stronger signal”, the page says. Access to detection “is granted by application”, so it is not a public checker. xAI says watermarking “doesn’t change response quality or speed”.
Opting out moves the legal duty to the customer
Only a team administrator can turn the mark off. The switch, labelled “Model output watermarking” in the Console’s team settings, works model by model and applies to every API key on the team, with changes reaching new requests “within 5 minutes”. Each change is recorded in the team’s audit log.
The opt-out comes with a declaration. The administrator must 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”. The page also warns that “some models may be locked on for your team”.
Why the Grok API watermark matters beyond Europe
xAI’s documentation names the EU AI Act but not a specific article. TeslaNorth linked the change to the Act’s transparency rules, which the European Commission explains in its Article 50 FAQ. Under Article 50(2), providers of systems that generate synthetic audio, images, video or text must make sure the output is “marked in a machine-readable format and detectable as artificially generated or manipulated”. The Commission says the obligation has applied since 2 August 2026, with systems already on the market before then given until 2 December to add marking. Fines can reach €15m or 3% of worldwide turnover.
The notable choice is scope. xAI has not limited the mark to European traffic. By putting it on every endpoint and making the customer opt out, it has set the EU rule as its global default and handed the legal judgment to any developer who wants unmarked text. That differs from OpenAI, which TSN reported is adding its textGrain watermark to ChatGPT and Codex text in the EU.
The limits are the same for every text watermark. The Commission’s own FAQ leaves short sequences and source code outside the marking duty, and a statistical mark needs enough text to show up. As with Google’s SynthID Detector, a “not watermarked” or “uncertain” result does not prove a human wrote something. And because detection is by application only, teachers, newsrooms and platforms cannot yet run their own checks unless xAI lets them.
For developers, the practical step comes first: check the x-watermark header after 7 November and decide whether an exception genuinely applies before touching the switch. More on xAI’s year of change is in TSN’s SpaceXAI timeline, and wider coverage sits on the AI regulation hub and Musk companies hub.
What to watch
The watermark switches on for the seven models on 7 November. The next questions are who xAI grants detection access to, and whether it says anything about marking in the consumer Grok apps before the Commission’s 2 December deadline for older systems.
Sources
- xAI (SpaceXSI) developer documentation: Output Watermarking, last updated 31 October 2026
- European Commission: Transparency obligations under Article 50 AI Act (FAQ), updated 24 July 2026
- European Commission: Code of Practice on marking and labelling of AI-generated content
- TeslaNorth: Grok’s API Answers Will Carry a Hidden Watermark Starting in November, 31 October 2026
- TokenPost: xAI Will Add Invisible Watermarks to Grok API Text Nov. 7, 31 October 2026
Related stories
- OpenAI Adds Invisible Watermarks to ChatGPT and Codex Text in the EU
- Google Opens Its SynthID Detector to Everyone: What It Can and Can’t Tell You
- xAI (now SpaceXAI): A Year From Grok 4.1 to the SpaceX IPO
Image: Colin / Wikimedia Commons, CC BY-SA 4.0, modified by TSN; this image is also CC BY-SA 4.0
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
