Starting November 7, text from the Grok API will carry an invisible watermark by default, according to a new page in SpaceXSI’s developer docs. The company says it’s doing this to comply with the EU AI Act.
The details
- It covers grok-4.7, 4.6, 4.5, 4.3 and the grok-4.20 reasoning, non-reasoning and multi-agent models, on every endpoint, for both the Responses and Chat Completions APIs.
- No hidden characters are added. The model slightly shapes which tokens it picks, and detection is statistical. Longer text gives a stronger signal.
- Check the
x-watermarkresponse header.truemeans the output was watermarked. - Team admins can opt out per model in the console, but only by confirming a legal exception applies. Some models may be locked on, and changes go into the audit log.
Tesla News Wire notes Anthropic and OpenAI made similar moves earlier this year. If you use Grok output in a pipeline, nothing breaks. Just know it’s marked.
God Bless, and May the Source Be with You!
