Langdon Winner uses the concept of “technological somnambulism” to criticize how we often accept technology-driven changes without fully questioning their deeper impacts. In recent years, there has been a strong push to develop responsible technologies, especially in the age of Artificial Intelligence (AI), that anticipate socio-ethical concerns and integrate human values into technological design. At first glance, this suggests that we are no longer “sleepwalking” through technological change but are instead taking deliberate steps to ensure that AI aligns with social values. However, this paper argues that Winner’s warning about technological somnambulism remains relevant today. Drawing on a multi-level framework of AI ethics, the paper shows how current efforts toward responsible AI often focus narrowly on anticipating and addressing impacts at the artifact level, while failing to challenge the deeper structural forces that perpetuate power imbalances and further marginalize vulnerable groups. This neglect of structural impacts risks leading us into what I call “AI somnambulism”. To move beyond this, the paper calls for strengthening existing ethical frameworks with a “structural imagination” to shift our focus from specific artifacts to deeper, structural issues in AI. It concludes by outlining how such imagination can be translated into more practical design practices, aiming to foster a truly responsible AI future.