xAI says that starting November 7 it will add text watermarks by default worldwide to Grok API.

Article

xAI Did not sign the EU transparency code,Grok API yet will add text watermarks by default worldwide

Beating AI News reports that xAI updated its developer documentation, announcing that starting November 7, it will add invisible text watermarks by default to the outputs of Grok 4.7、4.6、4.5、4.3 and three Grok 4.20 models.API The coverage includes not only the EU but also global and regional API; even structured outputs and tool-call parameters will be marked. It remains unclear whether the Grok web version and the built-in chat on X will be enabled at the same time.

This watermark works by slightly adjusting word-selection probabilities as the model generates text, leaving an identifiable statistical pattern across the passage. Direct copying and pasting will not remove the watermark, but shortening the text or substantially rewriting it will reduce detection reliability.

Behind this is the transparency requirement of the EU AI Act, with the relevant provisions taking effect on August 2. Although xAI did not sign the EU's voluntary transparency code, it must still comply with applicable legal obligations. Developer teams may disable the watermark on a per-model basis, but must confirm that the relevant legal exemptions apply and assume responsibility for marking and disclosure.

Google began using SynthID text watermarks for 2024 in Gemini; Anthropic began enabling watermarks globally for new Claude models this year August. OpenAI announced in October that it would gradually add watermarks to ChatGPT and Codex in the EU, while allowing users globally to API enable them; they remain disabled by default.

Original link https://m.theblockbeats.info/flash/374856

Key points

Beating reports that xAI updated its developer documentation, announcing that starting November 7 it will add invisible text watermarks by default to Grok 4.7、4.6、4.5、4.3 and three Grok 4.20 models'API outputs.
The watermark covers global and regional API, and structured outputs and tool-call parameters will also be marked.
Direct copying and pasting will not remove the watermark, but shortening the text or substantially rewriting it will reduce detection reliability.
Developers may disable the watermark on a per-model basis, but must confirm that the relevant legal exemptions apply and assume responsibility for marking and disclosure.
It remains unclear whether the Grok web version and the built-in chat on X will be enabled at the same time.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

This watermark is not a visible pattern; instead, it slightly adjusts word-selection probabilities, leaving the text with an identifiable statistical pattern.

Why it matters to readers

Using Grok API to generate content may require teams to address marking, disclosure, and applicable legal exemptions.

If you use Grok API to generate text, the default output may contain an unobtrusive watermark; copying the text will generally not remove it.

Risks and unknowns

The evidence provided does not include the original link to the xAI developer documentation, so the specific settings still need to be verified.
It remains unclear whether the Grok web version and the built-in chat on X use watermarks.
The specific legal exemption conditions and disclosure requirements for developers disabling the watermark remain unclear.

xAI says that starting November 7 it will add text watermarks by default worldwide to Grok API.

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center