SpaceXAI has announced that, starting November 7, it will add invisible watermarks to text generated by its Grok series models. The watermark is embedded in the token selection process during text generation and does not rely on hidden characters or metadata. Watermarking has been enabled for SpaceXAI APIs worldwide, covering multiple Grok models, and team administrators can disable the watermark for individual models. This change is intended to comply with the requirements of the EU Artificial Intelligence Act. Currently, only ChatGPT in North America has yet to introduce invisible watermarks for text.