everyone is putting AI in schools......

Anthropic’s Claude is rolling out an education-focused “learning mode” designed to do more than generate answers—aiming to push students toward critical thinking through Socratic-style questioning. The pitch is straightforward: instead of handing over solutions, Claude asks students “why” and presses for supporting evidence, such as what facts back a conclusion. That shift matters because AI is already being used in school at scale, and many students may be mistaking AI-assisted work for real learning.

Access to Claude’s education features appears tied to institutional agreements. Anthropic has been working with universities including Northeastern University, the London School of Economics and Political Science, and Champlain College, with “full campus access agreements” that provide Claude to students through campus arrangements. For individuals, the learning mode doesn’t seem broadly available; instead, students may need to participate through campus ambassador programs or apply for API credits to build projects. Beyond universities, Anthropic is also partnering with education technology vendors such as Instructure, whose Canvas learning management system is widely used to deliver coursework—meaning Claude could be integrated directly into existing teaching workflows.

The broader context is that AI in education is no longer hypothetical. A cited survey found 82% of undergraduates and 72% of K12 students have used AI for school, and 45% use it for tasks like writing assignments. Students often don’t view that help as cheating, treating it like Google—something that provides assistance on tricky problems. But research and reporting highlighted in the transcript suggest a risk: students can develop an “illusion” of understanding. Ethan Mollick, a professor at the Wharton School of the University of Pennsylvania, is quoted arguing that the key question isn’t whether AI will change education, but how society will shape that change to make learning more effective, equitable, and engaging.

That illusion is backed by an example from a high school in Turkey where students received GPT-4 help on homework. Homework scores improved, yet final exam performance dropped by 17%, suggesting that AI support may boost short-term outputs while weakening durable learning.

Another complication is detection. Teachers may believe they can spot AI-written work, but the transcript argues that detection is unreliable because AI text can be made to sound human with the right prompts—and “AI detectors” can’t reliably catch everything. With that reality, the proposed direction is less about policing and more about steering: if students will use AI anyway, education systems should encourage AI use that strengthens reasoning rather than replacing it.

For families outside large institutions, the transcript offers a workaround: using a carefully designed “tutoring prompt” that forces an AI tutor to ask the student questions, assess what they already know, and build understanding through open-ended prompts—mirroring Socratic questioning. The creator describes hosting a restricted AI environment (via Open WebUI) and limiting access so children can only use tutor-style models, not general-purpose tools. The underlying message is caution without denial: AI is coming to classrooms and homes regardless, so the goal is to guard against illusory knowledge and use AI to deepen thinking instead of substituting for it.