SpaceXSI, SpaceX’s AI unit, will add an invisible watermark to text from seven Grok models on its API from 7 November 2026, on by default for every team, according to its developer documentation. Its docs say it watermarks text output “to comply with the EU AI Act”.
That is 25 days before 2 December 2026, the date by which the amended AI Act gives providers of systems already on the market before 2 August 2026 to comply with its text-marking rule, by our arithmetic. OpenAI took the other route a month earlier: watermarking in its API stays off unless customers opt in.
Business Pill · THE MODEL AND THE HARNESS
A short explainer on the difference between an AI model and the harness around it. It teaches the general idea only and says nothing about this story.
The key insight: As we read the docs, SpaceXSI puts the compliance default on the provider side: the watermark is on for every team from 7 November 2026, and a team that opts out must confirm a legal exception applies and take responsibility for any marking, while OpenAI’s API watermark stays off unless customers opt in.
In an email shown in a screenshot posted on X by @ai_for_success, SpaceXSI wrote: “We’ve added support for text watermarking to SpaceXSI API outputs”. It listed grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three grok-4.20 variants: reasoning, non-reasoning and multi-agent.
The email says watermarking “is on by default for all teams and will be enabled on November 7, 2026”. It adds: “If you believe an exception under the law applies to your use, you can opt out per model from the SpaceXSI Console”.
Text-watermark defaults in the OpenAI and SpaceXSI APIs against the EU AI Act timeline: OpenAI opt-in from 5 October 2026, SpaceXSI on by default from 7 November 2026, and the 2 December 2026 Article 50(2) deadline for systems on the market before 2 August 2026. Days are our arithmetic.
How the Watermark Works
The docs say the watermark “is not visible”. It “slightly shapes which tokens the model picks”, so it “adds no characters, markup, or hidden text to the response”. There is no request parameter; a response header, “x-watermark: true”, shows that the output was generated with the watermark.
Detection is statistical. The docs say SpaceXSI “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”.
It covers ordinary text, structured outputs and tool-call arguments, on every SpaceXSI endpoint, including the global endpoint and regional ones, the docs say.
Opting Out Has Conditions
Team admins can switch the watermark off per model in the SpaceXSI Console, and the setting applies to every API key on the team. Opting out “asks you 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”, the docs say.
The docs also say “Some models may be locked on for your team”, and that every opt-in and opt-out is recorded in the team’s audit log.
The Rule Behind It
Article 50(2) of the EU AI Act, as shown in the AI Act Explorer, says providers of AI systems generating synthetic text and other content “shall ensure that the outputs of the AI system” are “marked in a machine-readable format and detectable as artificially generated or manipulated”.
Article 111(4), added by an amendment according to the Explorer, says providers whose systems were “placed on the market before 2 August 2026” must comply with Article 50(2) “by 2 December 2026”. The @ai_for_success post linked SpaceXSI’s change to Article 50(2); SpaceXSI’s docs cite the AI Act without naming the article.
The Business Lever: Defaults
SpaceXSI and OpenAI have chosen opposite defaults for text watermarks in their APIs. SpaceXSI turns it on for everyone and asks customers who switch it off to take on the legal responsibility. OpenAI, in its 5 October announcement, kept the API setting off by default and said it was not a global default at launch.
As we read SpaceXSI’s docs, a customer that opts out confirms an exception under applicable law applies and is solely responsible for any marking, while a customer that does nothing gets the watermark from SpaceXSI. Related: the shared AI-training opt-out the large labs backed this week, and France’s proposed data tax.
The Structural Read
As we read the docs, the change needs no code from developers: there is no request parameter, and only a response header, x-watermark, shows whether a response was watermarked.
As we read the docs, the watermark is statistical and detection is gated: SpaceXSI scores text and access to its detection API is granted by application.
As we read the AI Act Explorer, the amended Article 111(4) gives providers of systems already on the market until 2 December 2026, which SpaceXSI’s 7 November date precedes by 25 days, by our arithmetic.
SpaceXSI developer docs, Output Watermarking
“SpaceXSI watermarks text model output to comply with the EU AI Act”
Three Implications
FOR DEVELOPERS ON THE API The docs say opting out applies to every API key on a team and is recorded in the team’s audit log.
FOR RIVAL MODEL PROVIDERSOpenAI’s API watermark, announced on 5 October 2026, stays off unless customers opt in, our earlier report shows.
FOR EU COMPLIANCE TEAMS Article 50(2) asks providers to mark synthetic text in a machine-readable format and make it detectable, per the AI Act Explorer.
The framework says “The limiting factors are organizational readiness, governance, workflow redesign, training, and trust.”
As we read it, a watermark that is on by default is one way a model provider takes on part of the governance work for its customers. The framework is the owner’s argument and does not discuss SpaceXSI’s watermark.
The email is known from a screenshot posted by an X user; the same terms appear in SpaceXSI’s public docs, read on 31 October 2026. The post’s claim that every future model will be watermarked at launch is not in the email or the docs we read. SpaceXSI does not say how robust the watermark is to editing or paraphrase, and the AI Act’s marking rule has exceptions this piece does not assess.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
From 7 November 2026, text from seven Grok models on SpaceXSI’s API carries an invisible watermark unless a team admin opts out and accepts the legal responsibility, 25 days before the AI Act’s 2 December deadline for systems already on the market, by our arithmetic.
A note on sourcing. Facts come from SpaceXSI’s public developer docs, a SpaceXSI customer email shown in a screenshot posted on X, the EU AI Act text in the AI Act Explorer, and our 5 October report on OpenAI, all read on 31 October 2026. Day counts are our arithmetic. We did not contact SpaceXSI or OpenAI.
Short explainer · Grok text watermark on by default from 7 Nov
A vertical news Short summarising this story. Figures come from SpaceXSI’s docs and customer email and the EU AI Act text; our arithmetic is labelled.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
