AI-written text has started carrying an invisible signature, and regulators got the reader before you did. The EU AI Act's transparency code took effect on 2 August. It requires providers to mark AI-generated

…see more

text so other systems can identify it. Anthropic has started doing exactly that. Models released since that date carry an invisible statistical watermark — a variant of the SynthID-Text method Google DeepMind published in Nature in 2024. It nudges inconsequential word choices into a pattern only a key can decode. It survives copy and paste. It survives light editing. A full rewrite removes it. The detection API is in private preview, and the list of who gets it is the part worth reading twice: regulators, law enforcement, media, fact-checkers, researchers, and enterprises carrying their own compliance obligations. Notice what that isn't. It is not a browser extension for your competitor. It is an institutional capability. And it is deliberately imperfect in both directions. Anthropic says so plainly: a detected mark is not conclusive proof Claude wrote something, and the absence of a mark does not mean no AI was involved. So the era we are walking into is not clean detection. It is credible accusation that you cannot cleanly refute. You will not win that with a detector score. You win it, or lose it, on whether you decided your disclosure position before anyone asked. Decide it now, while it is cheap and boring. What do we use AI for. What does a named human sign off. What do we say if a customer or a journalist asks. Most brands will improvise that answer under pressure, badly, in public. If a customer asked tomorrow which parts of your marketing a human actually wrote, could you answer without going to check? #EUAIAct #ContentOps #BrandTrust