Generative artificial intelligence and physics education: what is changing, and how fast

Generative artificial intelligence solves and “explains” university-level physics, and its capabilities are improving in a regular, predictable way. This undermines an assumption on which much of our teaching rests: that a correct answer is evidence of reasoning and understanding. In this paper we argue that the response to this rupture is not a new technology, but the practices that physics education research (PER) has advocated for decades: a focus on process, inquiry-based laboratory work, and the assessment of reasoning. One disposition becomes even more essential: epistemic vigilance, that is, treating AI outputs as hypotheses to be tested. We discuss what AI can do in physics today, what its scores actually measure, what we risk losing, and what, in concrete terms, is worth learning for those who teach and those who learn.