After reading the whole article I still came away with the suspicion that this is a PR piece that is designed to head-off strict controls on LLM usage in education. There is a fundamental problem here beyond cheating (which is mentioned, to their credit, albeit little discussed). Some academic topics are only learned through sustained, even painful, sessions where attention has to be fully devoted, where the feeling of being "stuck" has to be endured, and where the brain is given space and time to do the real work of synthesizing, abstracting, and learning, or, in short, thinking. The prompt-chains where students are asking "show your work" and "explain" can be interpreted as the kind of back-and-forth that you'd hear between a student and a teacher, but they could also just be evidence of higher forms of "cheating". If students are not really working through the exercises at the end of each chapter, but instead offloading the task to an LLM, then we're going to have a serious competency issue. Nobody ever actually learns anything.
Even in self-study, where the solutions are at the back of the text, we've probably all had the temptation to give up and just flip to the answer. Anthropic would be more responsible to admit that the solution manual to every text ever made is now instantly and freely available. This has to fundamentally change pedagogy. No discipline is safe, not even those like music where you might think the end performance is the main thing (imagine a promising, even great, performer who cheats themselves in the education process by offloading any difficult work in their music theory class to an AI, coming away learning essentially nothing).
P.S. There is also the issue of grading on a curve in the current "interim" period where this is all new. Assume a lazy professor, or one refusing to adopt any new kind of teaching/grading method: the "honest" students have no incentive to do it the hard way when half the class is going to cheat.