Anthropic says it now embeds watermarks in AI-generated text and files, partly to meet the EU AI Act. The watermark uses word and phrase patterns rather than hidden characters, so it does not change output quality or token costs. Editing the text reduces or removes the mark, and AI-written code carries very little. A future API will let others check whether text came from AI. For a Thai business, there is no action needed now; if you publish AI-drafted content, know that it may be detectable.