With the implementation of the EU AI Act, AI companies and other interested observers will be able to identify text generated by AI applications. The law requires all companies to fully comply by December 2 by including a hidden watermark within all generated text. This watermark is not a letter or symbol visible to the naked eye, but rather a subtle mathematical and statistical pattern in the way words are chosen.
The generated text will appear completely natural, but approved scanning tools can accurately detect the watermark and confirm that it was generated by AI rather than written by a human. Consequently, it will become difficult to continue directly copying and pasting AI-generated text into school assignments, academic papers, official emails, or books, articles, and reports without it being identifiable as AI-generated.
The European Union issued the EU AI Act concerning the use and transfer of AI-generated text, with the transparency requirements under Article 50 coming into effect on August 2, 2026. The Act requires providers of generative AI systems to make their output machine-readable and detectable as AI-generated content. Anthropic, the company behind the Claude model, has already included a hidden watermark within its generated text, effective August 2, 2026. This watermark will be applied globally, not just to users in Europe, as there is currently no reliable way to restrict its application geographically. How does this technology work? AI companies embed invisible symbols or characters into each generated text.
The method works through a simple, subtle form of guidance during word selection. Instead of randomly selecting the next word from a reasonable range of possibilities, the model applies a slight statistical bias towards a specific set of “preferred” words, defined by a secret key. This allows detection tools to identify with confidence whether the text was generated by AI. Although the deadline for implementation is December 2, 2026, the three major companies have not yet fully committed, with the exception of Claude, owned by AI giant Anthropic.
All three companies have already signed the EU Code of Conduct on the Transparency of AI-Generated Content. The companies have stated their intention to implement similar technologies in the future. Currently, OpenAI only tags images, not text. The law only mandates compliance “whenever technically available,” meaning it is not 100 percent mandatory in all cases. There are ways to escape the AI trap.
By Ahmad alsarraf
email: [email protected]
email: [email protected]
