SpaceX Super Intelligence’s developer documentation schedules default-on invisible watermarking for text-generation output from seven Grok models starting November 7, 2026. The policy covers every SpaceXSI API endpoint, including the Responses API and Chat Completions API, and requires no request parameter or integration change, according to the official documentation.
Which Grok models and outputs are covered?
The scheduled policy names 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 applies to text-generation responses, including ordinary text, structured outputs, and tool-call arguments.
The watermark is a statistical signal created during generation by subtly shaping token choices. It adds no characters, markup, or hidden text to the response. The response format remains unchanged apart from the watermark header.
How can developers check a response?
For the Responses API and Chat Completions API, check the x-watermark response header. A value of true means that request’s output was watermarked; another value or no header means it was not.
A separate statistical detector can classify text as watermarked, not watermarked, or uncertain. The documentation says longer text produces a stronger signal, and access to the detection API is granted by application.
What can a team administrator change?
A team administrator can opt out selected models in the SpaceXSI Console, after confirming that an exception under applicable law applies and accepting responsibility for relevant marking, disclosure, or other legal requirements. The setting applies to every API key on the team. Changes affect new requests within five minutes, although some models may be locked on and unavailable for opt-out.
