Everything AI writes is now traceable (SSALAT L7FLA)

Everything Claude writes now carries an invisible watermark. It survives copy paste, you cannot see it, and it went live this month. If you use AI for anything at all, emails, reports, code, uni assignments, this concerns you directly. In this video I explain for everyone even non-technical how it works : An AI writes one word at a time. At each step it picks from a list of candidate words that all fit. Normally that pick is random. With watermarking, the randomness is replaced by a secret key combined with the last few words. The text still reads completely natural, but whoever holds the key can check whether your words match the sequence the key would have produced. One sentence proves nothing. A few paragraphs become a statistical fingerprint. No hidden characters, no visible tag, nothing you can find and delete. Sources: Anthropic, how Claude's text watermarking works https://www.anthropic.com/news/claude... Google DeepMind, SynthID Text paper published in Nature https://www.nature.com/articles/s4158... EU AI Act, Article 50, transparency obligations https://artificialintelligenceact.eu/... The watermark remover repo https://github.com/guillaumemeyer/wat... DeepMind's official SynthID Text implementation https://github.com/google-deepmind/sy...