SpaceXSI's API documentation says listed Grok models watermark text from 7 November

SpaceXSI's API documentation on output watermarking says it was last updated on 31 October 2026. The page says SpaceXSI watermarks text model output to comply with the EU AI Act, on every SpaceXSI API endpoint, including https://api.x.ai and any regional endpoint. It says that, starting 7 November 2026, responses from these models include a watermark, on by default, with an opt-out per model: 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.

The page says the watermark applies to text from the Responses API and the Chat Completions API, including ordinary text, structured outputs and tool-call arguments. It says the watermark is not visible and adds no characters. It says a response header x-watermark: true means that request's output was watermarked. It says team admins can turn watermarking off per model in the SpaceXSI Console, that the setting applies to every API key on the team, and that opting out asks the admin to confirm that an exception under applicable law applies. It says some models may be locked on and cannot be turned off from the Console. It says watermarking does not change response quality or speed, and that there is no request parameter.

What it means for business: the page describes a default text watermark on the listed API models, with a team-level opt-out that asks for a legal-exception confirmation. It does not publish a detection service that is open without an application. This describes the documentation. It is not advice on Article 50.