Article 50 duties to disclose AI interactions and label certain synthetic media are now enforceable, with the AI Office able to demand GPAI documentation, evaluate models and levy fines. Deepfake and chatbot disclosures apply day one; a narrow grace period to December covers only some machine-readable markers for systems already on the market. Roughly 190 organizations signed a related transparency code of practice. It matters for any provider whose outputs reach EU users. Caveat: early enforcement may ramp gradually even as legal exposure starts now.