Trustworthy AI through regulation? Sketching the European approach