_2025.png){kind=logo}
SpaceXSI, SpaceX's artificial intelligence unit formerly known as xAI, will switch on invisible text watermarking by default in the Grok API on November 7, 2026. According to the company's public documentation and reports published on October 31, 2026, the change applies to API-generated text and does not affect consumer-facing products.
The feature covers seven models: grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three grok-4.20 variants (reasoning, non-reasoning and multi-agent). It applies to all text-generation endpoints, including the main api.x.ai endpoint, the Responses API and the Chat Completions API, as well as structured outputs and tool-call arguments.
How the watermark works
The watermark is statistical and invisible: it gently steers the model's token choices toward a detectable pattern, without adding characters, markup or hidden text to the response. According to SpaceXSI, the process does not change the length, quality or speed of responses.
Detection is also statistical. The company says it scores a piece of text against the watermark and reports whether it is watermarked, not watermarked or uncertain, and that longer text gives a stronger signal. Access to the detection API is granted by application. In addition, the x-watermark response header indicates whether a response was generated with the watermark.
Restricted opt-out and regulatory context
Team administrators can switch the watermark off per model in the SpaceXSI Console, but the opt-out is framed narrowly: they must confirm that an exception under the law applies to their use and take sole responsibility for any marking, disclosure or legal requirement. The company also records every opt-in and opt-out in the team's audit log and says some models may be locked on for a team.
SpaceXSI's documentation says the aim is to comply with the EU AI Act, the European artificial intelligence law. Article 50(2) of the regulation requires outputs of systems generating synthetic text to be marked in a machine-readable format and detectable as artificially generated. According to analysts' reading of the amended rule, providers of systems already on the market before August 2, 2026 would have until December 2, 2026 to comply — so the November 7 date would precede that deadline by 25 days. Note that SpaceXSI's documentation cites the EU AI Act without naming the article.
A reversal of the earlier position
The change reverses the company's earlier public position, which argued that AI-generated text should remain unaltered. The new approach leaves the visible text unchanged but places the signal in the generation statistics.
The contrast with OpenAI is also notable: according to reports, text watermarking in OpenAI's API, announced on October 5, 2026, stays off by default and is applied only when the customer opts in. The announcement shared on X claims that, with xAI on board, ChatGPT would be left as the only North American model without text watermarking — that is a claim from the announcement, not an independent verification of every model on the market.
One sourcing caveat: the SpaceXSI customer email reached the public through a screenshot posted on X by the user @ai_for_success. The same terms, however, appear in the company's public documentation, read on October 31, 2026, which supports the central facts of this article.
Image credit: Grok AI (xAI) logo, available on Wikimedia Commons, authored by Grok AI (xAI), in the public domain as a text logo, with the trademark subject to trademark protection in some jurisdictions.
Sources: Crypto Briefing, FourWeekMBA and Wikimedia Commons.
Transparency note: this article was produced with the help of artificial intelligence, based on public posts on X, SpaceXSI's public documentation and reporting published on the internet on October 31, 2026. Readers are advised to consult the original sources to confirm technical details and deadlines.
Signature: Muse
_2025.png){kind=link}