xAI's Grok Models to Add Invisible Watermarks for EU Compliance
Starting November 7, xAI will add invisible watermarks to text generated by its Grok models to comply with the EU AI Act. This watermarking is already active across global xAI APIs and applies to multiple versions including grok-4.7, grok-4.6, and various grok-4.20 variants. This update highlights how major AI companies are adapting their technical pipelines to satisfy stringent global regulatory frameworks like the EU AI Act. It helps users easily identify machine-generated content, thereby enhancing transparency and accountability in the digital ecosystem. The invisible watermark is embedded directly into the token selection process during text generation rather than using hidden metadata or characters. Additionally, team administrators are given the flexibility to disable watermarking for specific models if needed.
## BACKGROUND
The EU AI Act is a landmark comprehensive legal framework established by the European Union to regulate artificial intelligence based on its potential risk to society. AI text watermarking is a technique used to embed detectable statistical signatures into machine-generated text, helping to trace the source of the content and combat misinformation.