Text generated by seven supported Grok API models now carries an invisible watermark by default. The provider, identified as SpaceXSI in its developer documentation, says the change took effect Nov. 7 and that it watermarks text model output “to comply with the EU AI Act.”
The covered models are grok-4.7, grok-4.6, grok-4.5 and grok-4.3, along with three grok-4.20 variants: reasoning, non-reasoning and multi-agent. The watermark is on across every SpaceXSI API endpoint, including the global api.x.ai endpoint and any regional ones, and it applies to both the Responses API and the Chat Completions API. Structured outputs and tool call arguments get the mark along with ordinary text responses.
No extra characters are added to the output. The watermark “slightly shapes which tokens the model picks,” the documentation says, so a response carries no extra characters, markup or hidden text. Detection is statistical. SpaceXSI scores a passage and reports whether it is watermarked, not watermarked or uncertain, and longer text gives a stronger signal. Access to the detection API is granted by application, and the company says watermarking doesn’t change response quality or speed.
Checking a response, and opting out
Existing integrations don’t need code changes. There’s no request parameter, and the response format is the same apart from a new x-watermark header. A value of true means that request’s output was watermarked; any other value, or no header at all, means it wasn’t. The documentation includes JavaScript, Python and curl examples for reading it, including the OpenAI SDKs pointed at SpaceXSI’s base URL.
Opting out is a team-level decision. Only team admins see the “Model output watermarking” row in the SpaceXSI Console, where they can switch the mark off model by model. The setting covers every API key on the team and reaches new requests within five minutes. Opting out asks admins to confirm “that an exception under applicable law applies to your use, and that you are solely responsible for any marking, disclosure, or other legal requirement.” Some models may be locked on and can’t be turned off from the Console, and each opt-in and opt-out is recorded in the team’s audit log.
How OpenAI and Anthropic set their defaults
The default puts SpaceXSI’s API on different footing from OpenAI’s. When OpenAI laid out its approach on Oct. 5, it said API customers globally could opt in to text watermarking for select models, but that the feature “will remain off by default in the API.” ChatGPT and Codex text in the European Union was set to get the mark over the following weeks, a plan WebProNews covered at the time.
Anthropic also applies watermarking worldwide. It said in August that future Claude models would carry a watermark based on Google DeepMind’s SynthID-Text method, which it described in a post as one that adds no hidden characters. Its help center says marks apply to output from supported models across the Claude API and apps “wherever Claude is offered, worldwide.”
Detection stays gated at all three companies. OpenAI limits its text detector to approved researchers and expert organizations, Anthropic’s detector is in private preview for eligible organizations, and SpaceXSI takes requests for its detection API through a form. Google’s public SynthID Detector, opened in October, checks images, video and audio files.
WebProNews is an iEntry Publication