xAI to add invisible watermarks to Grok API text output by default from Nov 7

The default-on, opt-out design is the telling detail: xAI is shifting the burden of provenance onto administrators who must actively disable it per model, rather than asking anyone to turn it on. Because the watermark extends to structured outputs and tool-call parameters, it reaches machine-to-machine pipelines, not just prose, which is where downstream systems may parse or relay Grok-generated text without knowing it carries a marker. Whether the watermark survives round-tripping through other tools, and how many administrators actually disable it, are the open questions worth watching.