xAI has not signed the EU transparency code, yet the Grok API will add text watermarks by default globally.

Beating AI News Flash: xAI has updated its developer documentation, announcing that starting November 7, invisible text watermarks will be added by default to API outputs for Grok 4.7, 4.6, 4.5, 4.3, and three Grok 4.20 models. The coverage extends beyond the EU to global and regional APIs, and even structured outputs and tool-calling parameters will be marked. It is not yet clear whether Grok's web version and the built-in chat on X will enable this simultaneously.

This watermark works by slightly adjusting word-selection probabilities during model text generation, leaving a recognizable statistical pattern across the entire passage. Direct copy-pasting will not remove the watermark, but shortening the text or substantially rewriting it will reduce detection reliability.

This is driven by the transparency requirements of the EU AI Act, whose relevant provisions took effect on August 2. Although xAI has not signed the EU's voluntary transparency code, it must still comply with applicable legal obligations. Developer teams can disable the watermark on a per-model basis, but must confirm compliance with relevant legal exemptions and bear the responsibility for marking and disclosure themselves.

Google already applied SynthID text watermarking to Gemini in 2024, and Anthropic began enabling watermarks globally for new Claude models in August this year. OpenAI announced in October that it would gradually add watermarks to ChatGPT and Codex in the EU, while allowing global API users to enable them voluntarily, with the default remaining off.