Grok API text watermarking is now 'off' be default (docs.x.ai)

Grok has announced that its text watermarking feature will be 'off' by default, allowing teams to enable it per model as needed. This watermarking embeds an imperceptible signal in generated text and audio, aiding compliance with the EU AI Act by ensuring outputs can be identified as AI-created. The feature is applicable to all text generation APIs, including Responses API and Chat Completions API, and once activated for a model, it automatically applies to all requests without requiring further coding. This development is significant for the AI/ML community as it reinforces transparency and accountability in the use of AI-generated content. The watermarking technique, based on SynthID Text from Google DeepMind, functions by subtly influencing token selection during text generation, leaving no visible alterations. Stakeholders can detect watermark status through an x-watermark header. With the rising regulatory scrutiny over AI outputs, Grok's approach supports organizations in adhering to legal standards and promotes ethical AI practices.