Artificial Intelligence (AI) is becoming increasingly important in medicine, with machine learning (ML) playing a crucial role in diagnosing and accurately predicting the progression of diseases. Even in psychiatry, promising developments based on ML are emerging. However, these advances also raise questions, such as the reliability of algorithm behaviour predictions. How dependable are these predictions? What happens if the AI predicts risky or aggressive behaviour? What standards should we set for these AI systems? Using concrete examples, we can explore the current capabilities, limitations, and ethical challenges posed by predictive AI in psychiatry. The presentation was part of the Next Generation Lecture Series organized by the Polytechnic Foundation of Frankfurt am Main.