SpaceXSI quietly flips Grok API text watermarking to off by default

SpaceXSI has reversed its decision on watermarking text generated by its Grok AI models. Initially, the company planned to automatically watermark text starting November 7, 2026, as a compliance measure for the EU AI Act. However, the feature is now off by default, with developers needing to enable it manually through their team’s administration settings. The change was made without public explanation, and it remains unclear if the original policy was ever implemented. This shift places the responsibility of compliance on developers, and SpaceXSI has become more cautious in its documentation about the sufficiency of watermarking to meet EU requirements.