xAI will watermark Grok API text by default from 7 November for the EU AI Act

What happened

From 7 November 2026, xAI adds an invisible statistical watermark to Grok API text by default to comply with the EU AI Act [1].

What changes

The watermark covers ordinary text responses, structured outputs and tool call arguments. Team admins can opt out per model, but must confirm that a legal exception applies to their use. Developers can check the x-watermark response header, where true means the output was watermarked.

Checked 2 Nov 2026

xAI will add an invisible watermark to text from its Grok API models by default from 7 November 2026, according to a new page in its developer documentation [1]. The company, branded SpaceXSI in the docs, says it is doing this to comply with the EU AI Act [1][2].

Which Grok models get the watermark?

The docs list seven models: grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three grok-4.20 variants for reasoning, non-reasoning and multi-agent use [1]. It applies to every text generation API, including the Responses API and the Chat Completions API [1].

The watermark covers ordinary text responses, structured outputs and tool call arguments. [1] It is enabled on the global endpoint and on regional endpoints [1].

How does the invisible watermark work?

The watermark is applied during generation by slightly shaping which tokens the model picks, so it adds no characters, markup or hidden text [1]. xAI says it does not change response quality or speed [1].

Detection is statistical: text is scored as watermarked, not watermarked or uncertain, and longer text gives a stronger signal [1]. Access to the detection API is granted by application [1].

Developers can check the x-watermark response header, where true means the output was watermarked. [1] No code changes are needed, the docs say [1].

Can developers opt out?

Team admins can opt out per model, but must confirm that a legal exception applies to their use. The setting applies to every API key on the team and reaches new requests within five minutes [1].

Some models may be locked on and cannot be switched off from the console, and every opt-in and opt-out is recorded in the team's audit log [1]. TeslaNorth was among the outlets that reported the change on 31 October [2].

Our read

Opinion: Watermarking by default is the cleanest way to meet the AI Act's marking rules, and xAI documents it plainly. But detection sits behind an application form, so for now the people best placed to check where a text came from are xAI and whoever it lets in.

What we could not verify

• How robust the watermark is against paraphrasing or translation; the docs do not say.
• Whether consumer Grok apps will also be watermarked; the page covers the API only.
• Which teams have models locked on, and why.