Grok API watermarks text by default from November 7

xAI will watermark text from its Grok API by default starting November 7, 2026. Its developer docs, which now carry the SpaceXSI name, say the change is to comply with the EU AI Act and applies to seven Grok models on every API endpoint.

What changed

• Covered 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 on the global endpoint https://api.x.ai and every regional endpoint, across the Responses API and the Chat Completions API. Ordinary text, structured outputs and tool call arguments are all marked.
• There is no request parameter and the response format does not change. A new x-watermark response header returns true when the output was watermarked.
• The mark is invisible. It shapes which tokens the model picks and adds no characters or markup. Detection is statistical, returns watermarked, not watermarked or uncertain, and is open only by application.
• Team admins can opt out per model in the console. The setting covers every API key on the team and applies to new requests within 5 minutes. Opting out means confirming that a legal exception applies to your use, and some models may be locked on.

What it means

From November 7, any Grok text you store, publish or pass to other systems, including JSON and tool arguments, carries a mark that xAI can detect. You cannot strip it with a text cleaner because nothing is added to the string. Teams that need unmarked output have to go through the opt-out and its legal confirmation.

Grok on RouterPlus

Grok rows in the RouterPlus catalog, November 2, 2026, per million tokens:

xAI's page covers requests to xAI's own API endpoints. It does not say whether output served through other hosts carries the mark. Compare every model on the RouterPlus models page.

What to do

Check the header on a direct xAI request after November 7:

curl -i https://api.x.ai/v1/responses \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "grok-4.7", "input": "Write a haiku about the ocean."}' | grep -i x-watermark

• Log x-watermark next to the model ID if you need an audit trail.
• If you plan to opt out, have the team admin and legal review it before November 7. The choice is per model, so check each Grok model you call.

Sources

1. SpaceXSI docs: Output Watermarking
2. RouterPlus model catalog