Differentiating human-written text from AI-generated text gets more and more complicated by the day, but in the European Union there are measures in place to prevent this, especially when transparency and legal obligations enter the game. Now Claude creator Anthropic says new models will actually mark AI-generated content from launch. Specifically, Claude models launched in the EU on or after 2 August 2026 will support machine-readable marking, with generated text carrying embedded watermarks and supported files carrying digitally signed provenance metadata. Anthropic claims it is also working to add marking to older Claude models.
Basically, this means that Claude-generated text will include an invisible watermark. Anthropic explains the watermark is woven into the text itself, does not affect meaning, quality or readability, and can travel with the text when it is copied and pasted elsewhere. However, concerns about the quality of the generated text or about potential tools and techniques to cheat the system have been risen already, for instance by GPTZero's CTO Alex Cui.
In fact, the system is designed for transparency, but it is not fool proof. Anthropic insists detection will indicate that content may have been processed by Claude, not prove full authorship. Marks may also disappear or become undetectable if text is heavily edited, paraphrased, translated, mixed with other writing, or if a passage is too short.
Do you generate text with AI or polish your own writing already? With what model?
