Nov 7, 2026 · found Nov 9, 2026, 12:30 PM PT
xAI turns on invisible text watermarking in the Grok API by default
Verdict: Confirmed · Model releases · AI security
xAI's docs say text from seven Grok models is statistically watermarked from 7 November unless a team admin opts out; how reliable detection is has not been tested outside xAI.
xAI documents a default-on watermark for text generated through its API.
- Claim: From 7 November 2026, text output from grok-4.20 through grok-4.7, including structured output and tool-call arguments, carries an invisible watermark. xAI docs
- Evidence: The watermark shapes token choice rather than adding characters. Detection is statistical and returns watermarked, not watermarked or uncertain, and responses carry an x-watermark header. xAI docs
- Caveats: Short text gives weak signals, detection API access is by application, and some models are locked on for certain teams. The docs say there is no quality or speed effect; nobody outside xAI has measured that. xAI docs
- Why it matters: Team admins can opt out per model after confirming a legal exception. The page ties the change to the EU AI Act. xAI docs
Verdict history
- Nov 9, 2026, 12:30 PM PT: Confirmed. Docs page read. The existence and terms of the change are confirmed; the detection accuracy has no independent test. Docs name the vendor as SpaceXSI in the console wording.