From Nov. 7, 2026, text generated through the Grok API will include an invisible watermark by default, according to a new page in SpaceXSI's developer documentation. SpaceXSI says the watermark lets it later recognize text as model-generated and is meant to comply with the EU AI Act.
Models covered
- grok-4.7, grok-4.6, grok-4.5 and grok-4.3
- grok-4.20-reasoning, grok-4.20-non-reasoning and grok-4.20-multi-agent
It applies on every endpoint, including the global api.x.ai and regional endpoints, and to both the Responses and Chat Completions APIs. Ordinary text, structured outputs and tool-call arguments are all covered.
How it works
The docs say the mark adds no characters or hidden text. Instead, it slightly shapes which tokens the model picks. Detection is statistical and returns watermarked, not watermarked or uncertain, with longer text giving a stronger signal. Access to the detection API is by application.
explainx.ai notes that Grok follows similar watermarking moves by Anthropic and OpenAI earlier in 2026.
Developers can check the x-watermark response header; a value of true means the output was watermarked. No code change is needed, and SpaceXSI says quality and speed are unaffected.
Opting out
Team admins can turn watermarking off per model in the SpaceXSI Console. The change applies to every API key on the team within 5 minutes. Opting out requires confirming that a legal exception applies to your use. Some models may be locked on, and every change is recorded in the team audit log.
