xAI Grok API to Default Invisible Watermarks from Nov 7, Admins Can Opt Out per Model
Frontier AI models · AI coding · Enterprise AI
xAI updated developer documentation stating that starting November 7, 2026, text generated by 7 Grok models via API will include invisible watermarks by default to comply with the EU AI Act. The change applies globally, and only team admins can opt out per model.
According to xAI's updated developer documentation, starting November 7, 2026, text generated by 7 Grok models—including grok-4.7 and grok-4.6—via API will default to embedding invisible watermarks to meet EU AI Act Article 50 requirements. The policy applies globally, covering standard text, structured outputs, and tool call parameters. Developers can check the 'x-watermark' response header to verify status. xAI explained that the watermark is embedded by subtly adjusting token selection without adding characters, with detection APIs available via application. Only team admins can opt out per model and must accept legal responsibility for labeling.
Key takeaways
- Starting November 7, 2026, xAI defaults invisible watermarking across 7 Grok model APIs globally, including tool call parameters.
- The watermark adjusts token selection without adding characters; developers can check via the x-watermark header, while detection API requires application.
- Only team administrators can opt out per model via dashboard settings, taking effect within 5 minutes under administrative legal liability.
Source：INSIDE Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human. Translation checked by AI.