SpaceXSI API leaves text watermarking off by default, official docs show; team admins can turn it on, alleged notice email says "as of Nov. 9," Musk says only "where required by law or requested by the user"

Official docs say text watermarking is "off by default"

SpaceXSI, the developer of the AI model "Grok," says in its official developer documentation that "Text watermarking is off by default" for text generated through its API. To add the watermark, a team admin turns it on per model in the team settings of the "SpaceXSI Console."

Watermarking embeds an invisible signal in text and audio generated by Grok models so the output can later be identified as AI-generated. The documentation says it can help enable compliance with the European Union's AI Act (EU AI Act). Text watermarking uses "SynthID Text," published by Google DeepMind in the journal Nature in 2024. Rather than adding characters or symbols, it slightly shapes which tokens the model picks, and whether a watermark was applied can be checked through the x-watermark header in the response. Access to watermark detection is granted by application.

Setting applies per team; turning it off requires a legal confirmation

According to the documentation, the text models that can include a watermark are "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). It applies to every text generation API, including the Responses API and the Chat Completions API. The setting belongs to the team and covers every API key on the team. Changes apply to new requests within 5 minutes, and every opt-in and opt-out is recorded in the audit log.

Turning a model off asks the user to confirm that an exception under applicable law applies to their use, and that they are solely responsible for any marking, disclosure or other legal requirement. The documentation says some models may be locked on for a team and cannot be turned off from the console.

Audio is handled differently from text. The documentation says audio from the voice APIs is watermarked by default, and while team admins can opt a voice model out, audio is still watermarked when the end user is in the EU, the European Economic Area (EEA) or California.

Alleged notice email says "as of November 9"

X user @techdevnotes (Tech Dev Notes) posted on Nov. 13 (the morning of Nov. 14 Japan time) that text watermarking is now off by default in the SpaceXSI API. The account is not an official SpaceXSI account. The post includes one image that appears to show an email.

In the image, the sender is "SpaceXSI <updates@email.api.x.ai>" and the subject line matches the post. The body says that as of November 9, 2026, text outputs are watermarked only if a team admin turns it on, and that watermarking can be enabled for all text models available via the API. It says no action is needed for those who do not want watermarking, including teams that already opted out, and that those who want it need a team admin to turn it on as described in the documentation. It says users are expected to review their own compliance with laws, and that those required to mark AI-generated text should turn watermarking on for the models they use.

ORIENS has not been able to independently verify whether the email is genuine. The "November 9" date in the email does not appear on the watermarking page of the official documentation or in the release notes. The release notes do not list the change to watermarking itself. The image does not show which pages the "our docs" and support contact links lead to.

Musk says "only on where required by law or requested by the user"

Elon Musk quoted the @techdevnotes post on Nov. 14 (early Nov. 15 Japan time) and wrote that text watermarking with Grok is "only on where required by law or requested by the user."

Musk's post does not mention the date in the email, the setting by team admins or the models covered. The official documentation describes text watermarking as something a team admin turns on, and does not say it is turned on automatically where the law requires it. The closest wording to watermarking being applied where the law requires it is in the documentation's description of audio watermarking based on the user's location.

The official documentation and release notes do not explain why the watermark was made off by default, or any effect on pricing or regions.