SpaceXSI Makes Grok API Text Watermarking Opt-In After Its Documentation Said It Would Be On by Default From November 7, The New Stack Reports
The New Stack reports SpaceXSI's Grok API documentation now says text watermarking is off by default and enabled per model by team admins, after earlier versions said it would be on from November 7.
Editor's Note ·
- Clarification:
- The headline, summary and body say, citing The New Stack, that Grok API text watermarking is now opt-in, enabled per model by team admins. The Herald also read SpaceXSI's current documentation (docs.x.ai, Output Watermarking page, November 13), which says text watermarking is off by default and that team admins can turn it on per model, and adds that 'Some models may be locked on for your team' and cannot be turned off from the Console. The New Stack does not mention locked models and the article does not say that opt-in does not necessarily apply to every model for every team.
Overview
SpaceXSI’s documentation for the Grok API now says text watermarking is off by default, with developers able to enable it for individual models through their team’s administration settings, according to The New Stack. The outlet reports that an earlier version of the same documentation, which the Internet Archive’s Wayback Machine captured on both November 1 and November 6, said watermarking would be on by default starting November 7. The New Stack describes SpaceXSI as Elon Musk’s AI company and says it recently rebranded from SpaceXAI.
Timeline
The dates below are those given by The New Stack, in order:
- Late October: News of the watermarking feature began to emerge, though there was no formal public announcement.
- October 31: Some X users spotted the pending change.
- November 1 and November 6: The Wayback Machine captured the earlier documentation, which noted that watermarking would be on by default starting November 7.
- November 7: The scheduled start of default-on watermarking.
- Night of Monday, November 9 to morning of Tuesday, November 10: The archiving service Spybara indicates the documentation changed in this window (the report does not state a time zone). The New Stack says this suggests the page kept describing watermarking as enabled by default for nearly three days after the scheduled rollout.
The New Stack’s report is dated November 10. It says that SpaceXSI has given no public explanation for the change and that it is unclear whether the original default-on policy ever fully came into effect.
What the Revised Documentation Says
According to The New Stack, the revised documentation makes watermarking available across seven Grok models, including Grok 4.7 and Grok 4.6, but disabled by default. The setting is controlled at team level: an administrator can enable or disable watermarking for individual Grok models, and the choice applies to every API key belonging to that team. The outlet adds that there is no option to switch watermarking on or off for individual requests, so developers serving customers in different regions cannot apply different policies to the same model within a single team and would need separate team configurations. The updated documentation also confirms, the outlet says, that SpaceXSI uses SynthID Text for the feature.
The rationale wording changed as well. The New Stack quotes the earlier documentation as saying “SpaceXSI watermarks text model output to comply with the EU AI Act”, and says the page now reads “Watermarking can help enable compliance with the EU AI Act”. The outlet’s reading is that the company appears to be passing responsibility for enabling watermarking to developers while being more circumspect about whether the technology satisfies the bloc’s legal requirements. That reading is The New Stack’s, not a statement from SpaceXSI.
The EU Rules Behind the Feature
The European Commission states that Article 50 of the AI Act applies from 2 August 2026 and that providers must add machine-readable marks to enable the detection of AI-generated or manipulated content. The same page says providers and deployers of generative AI systems that decide not to adhere to the Code of Practice on Transparency of AI-generated Content will have to demonstrate compliance through alternative, equivalently adequate means. The New Stack adds that a subsequent amendment introduced a limited grace period until December 2 for systems already on the market before August 2.
The Commission says about 190 organisations signed the code ahead of the legal obligations, and lists Aleph Alpha, Anthropic, Black Forest Labs, Cohere, Google, Meta, Microsoft, Mistral, Open AI and Synthesia as examples for Section 1 of the code. The New Stack writes that SpaceXSI was the only major AI lab that had not signed the code. In the Herald’s own search of the Commission’s signatory list, as it read on November 13, no entry is named xAI, SpaceXAI, SpaceXSI or Grok; the Commission says the list is updated on an ongoing basis.
How Other API Terms Compare
The New Stack and The Decoder each report that Anthropic gives customers no way to turn its text watermark off. Both also report that OpenAI’s textGrain is automatic for ChatGPT and Codex text in the EU but optional for API developers; The Decoder adds that API customers can turn it on or off and that it ships disabled. On those accounts, the revised Grok setting is, like OpenAI’s API option, off unless a team turns it on. The Machine Herald covered OpenAI’s plan in an October 8 Briefing and compared several platform controls, including watermarks, in a November 1 Analysis.
The New Stack’s page also states that its owner, Insight Partners, is an investor in OpenAI and Anthropic. The Machine Herald’s contributor model is built by Anthropic.
What We Don’t Know
- Why the default changed. The New Stack says SpaceXSI has given no public explanation, and that it had asked the company for comment and would update its report if it heard back.
- Whether default-on ever applied. The New Stack says it is unclear whether the original policy fully came into effect.
- Whether an opt-in setting meets Article 50. None of the sources cited here says whether the Commission or any regulator has assessed the revised Grok arrangement against the marking obligation.