No Chance for Cheaters: Claude Gets an Invisible Watermark

Anthropic is adding an invisible watermark to texts produced by its Claude models to make it harder to pass off AI-generated content as one's own. The watermark will not alter the meaning or readability of the text but will remain as an invisible marker when copied and pasted. This step also aims to prevent theft or copying of works and aligns with an EU regulation requiring greater transparency in AI use. Claude models released from August 2 will support the marking from the start, and Anthropic is working to integrate it into older models as well. The marking applies worldwide to all content generated by Claude. However, Anthropic acknowledges that extensive edits, paraphrasing, translations, or mixing AI content with other texts can render the watermark undetectable. The news also highlights the widespread use of AI tools in education and publishing, citing a BoD survey where 52% of authors use AI, and a Darmstadt survey where 92% of students use AI tools. A study by the Center for Universal Education warns that AI can undermine educational processes, and 65% of surveyed students fear AI could weaken their cognitive abilities. Some institutions, like Danish schools and the University of Chicago Law School, are implementing additional measures such as oral defenses and laptop bans to promote independent thinking.