SpaceXAI adds invisible watermarks to Grok model outputs in response to EU AI regulations

SpaceXAI will begin adding invisible watermarks to text generated by its Grok models starting on November 7, 2026. The watermarks are embedded during the token selection process and are not visible to the naked eye. The move is in response to the EU's Artificial Intelligence Act, which requires all AI systems to include traceability mechanisms. The watermarks will be applied to multiple Grok versions, including grok-4.7, grok-4.6, grok-4.5, grok-4.3, grok-4.20 reasoning, grok-4.20 non-reasoning, and grok-4.20 multi-agent. Administrators can opt to disable the watermark for specific models. AI governance observer Andrew Curran noted that no other model, including ChatGPT in North America, currently uses invisible watermarks for text.