Claude will hide a watermark in your AI-written text, and it’ll follow you around

Claude, the AI language model, will now embed invisible watermarks in the text it generates. These marks, set to debut on August 2, 2026, will be globally applicable and will even persist if the AI-written content is copied and pasted elsewhere. Moreover, these watermarks can survive certain editing attempts. Anthropic, the company behind Claude, is also planning to equip its existing model versions with this watermark feature during the EU AI Act's transition period.

This move is in line with Anthropic's commitment to the Article 50(2) Code of Practice on Transparency of AI-Generated Content under the EU AI Act. In a recent update, Anthropic disclosed that new Claude models will begin implementing these invisible watermarks from August 2, 2026, regardless of the EU regulation's influence. The impact of these markings will be felt globally, as long as Claude is accessible.