xAI plans default text watermarking for seven Grok API models
The AMW Read
Default watermarking changes how a major model provider marks API output, with implications for developers and content provenance.
xAI plans default text watermarking for seven Grok API models
Starting November 7, 2026, xAI plans to watermark text generated by seven Grok models through its APIs. The reported change covers the Responses and Chat Completions APIs, including structured outputs and tool-call arguments. Team administrators can disable marking by model after a legal confirmation. The source reports no announced date for the Grok app, website, or X subscription.
The move adds output provenance to competition among foundation model providers, with consequences for developers that distribute generated text. The watermark reportedly changes word-selection probabilities rather than adding visible characters or metadata to the text. Detection remains limited: xAI controls access to the detector, and short or heavily edited passages can produce uncertain results. The source says xAI's claims of no quality or speed impact lack independent testing.
Builders using Grok should check which models their applications call and record the x-watermark response header when they need evidence that a request used watermarking. They should also test structured outputs, tool calls, quality, and latency before and after the change. The header records the status of an API response; it does not give downstream readers independent access to a detector.
