Grok responses get an invisible watermark in the API starting Nov. 7

Developers who use Grok's API will get text with an invisible watermark by default starting Nov. 7, 2026. This is according to the company's own documentation. The company name appears as both xAI and SpaceXSI in the sources.

The API is the access point through which other companies and programmers plug Grok's language model into their own systems. The change is therefore aimed at developers and not directly at ordinary users. It is not documented whether the consumer app Grok or X is covered.

What is the watermark?

It is invisible and adds neither characters nor hidden text. Technically, the model slightly influences which word fragments (tokens) it picks as it writes, creating a statistical pattern. According to Tesla North, longer texts produce a stronger signal.

The marking applies to seven models: grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three versions of grok-4.20. It covers text from text-generation APIs, including structured responses and tool-call arguments.

According to the documentation, the watermark affects neither quality nor speed, and developers do not need to change any code. It has not been independently verified that it works as described.

Can you opt out?

Team administrators can turn off watermarking for individual models in the SpaceXSI Console. It takes effect within five minutes, and the actions are logged. Some models may, however, be locked to on. According to Tesla North, opting out requires the administrator to confirm that a legal exemption applies to the use.

A text cannot simply be read with the naked eye. Detection is statistical and returns one of the answers watermarked, not watermarked or uncertain. Access to the detection API requires an approved application. Each response also carries a header called x-watermark, which shows whether the response is marked.

Background: EU rules

The documentation states that text output is marked to comply with the EU's AI Act. According to Tesla North, Article 50 requires providers to mark synthetic text, images, video and audio in a machine-readable way. The rules took effect on Aug. 2, 2026.

In August, City AM reported that xAI was not among the signatories of the EU's voluntary code of practice on transparency for AI-generated content. Anthropic, OpenAI, Google, Microsoft, Meta and Mistral were. According to City AM, Grok still had to comply with the regulation's requirements. On Aug. 11, Anthropic announced plans for invisible, machine-readable watermarking of text from Claude.