xAI plans to turn on invisible text watermarking by default for seven Grok API models starting Nov. 7, 2026, according to its developer documentation. The listed models are 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. The planned coverage extends across global and regional API endpoints. It includes text generated through the Responses and Chat Completions APIs, whether that text appears in ordinary responses, structured outputs or tool-call arguments.
The watermark slightly influences which tokens a model chooses as it generates text; it adds no characters, markup or hidden text. xAI describes detection as statistical and says longer text gives a stronger signal. Access to its detection API is granted by application. For an individual Responses or Chat Completions API request, an x-watermark: true response header means the output was generated with the watermark. Any other value, or no header, means it was not.
Team admins can turn watermarking off for some or all models, with the setting applying to every API key on their team. Changes take effect for new requests within five minutes. Some models may be locked on for a team and cannot be turned off in the console. To opt out, an admin must confirm that an exception under applicable law applies to their use and accept responsibility for any marking, disclosure or other legal requirements.
