For the first time anywhere on Earth, the law now requires an AI to say hello properly. On August 2, the enforcement phase of Europe's AI Act switched on, and with it a simple rule with sharp teeth: if people are talking to a machine, the machine has to say so.
The new transparency regime covers chatbots and voice assistants (disclose that you're AI), deepfakes (label them), and AI-generated media generally (carry machine-readable marks so software can detect it). Ignore it and the penalty runs up to €15 million or 3 percent of worldwide turnover, whichever hurts more. Meanwhile the heavier “high-risk” rules — hiring tools, credit scoring, critical infrastructure — got pushed out to late 2027 and 2028, and Brussels added a flat ban on so-called nudifier apps.
Why should a solo builder in Philadelphia or Portland care about Brussels? Because the law follows the user, not the server. If your chatbot, your generator or your product's AI output reaches someone in the EU — and on the open internet it will — you're in scope.
The move: add the one line of honesty now. “You're chatting with an AI” costs nothing, and labeling generated media is an afternoon of work. Better yet, flip it into a feature — compliance-ready is now a phrase clients will pay for, and the builders who wired it in early get to charge for the head start.
