Grok now watermarks every API reply by default

Starting November 7, text from seven Grok models — grok-4.7 down through grok-4.3 and the grok-4.20 family — carries an invisible statistical watermark by default on every xAI API endpoint, with no code change required. The mark shapes token choice as the model writes, adds no characters or hidden text, and shows up as an x-watermark: true response header; detection itself is statistical and available by application only.

The stated reason is the EU AI Act: Article 50 requires machine-readable marking of AI-generated content ahead of a December 2 deadline. Opting out is deliberately friction-heavy — only team admins can do it, per model, and only after confirming a legal exception applies and the team takes sole responsibility for marking. It is the first default-on watermark policy at a major frontier API; Anthropic's second round of text watermarking starts November 16.