New research and a television experiment in Estonia show that AI-generated voices are becoming difficult to distinguish from real human speech, raising alarm among security experts about a potential surge in phone scams targeting Estonian-speaking populations.