Why is Claude for Teachers?

Anthropic bumbles its way into education

Everything about AI right now is weird.

I took the first two weeks of July off to road trip through New Mexico (highly recommended) and managed to avoid reading or thinking about AI for two weeks (also highly recommended). When I came back online, the email at the top of my inbox was from a reporter at Ed Week asking if I’d care to comment on “Claude for Teachers,” and my initial reaction was—huh?

For the last four years, OpenAI and Google and Meta have largely aimed their AI efforts at getting “consumers,” business-speak for “individual humans,” to use their AI products. It’s never been entirely clear that there’s a business model to be built around this, a fact that Ed Zitron reminds us via 50,000 word screeds every few weeks, but the basic play in technology for the past several decades has been to acquire users first and monetize later. And in accordance with this approach to lucrative enshittification, it makes sense for each of these companies to have an “education play,” meaning, a dedicated effort to infuse their particular product into the education system so it becomes the one that users, which is to say kids and their teachers, become habituated to using. This sort of thing.

Meanwhile, as these titans of commerce have battled it out, Anthropic went the other way. Instead of focusing on consumer usage, Anthropic quite deliberately focused on developing AI for “enterprises,” business-speak for “businesses.” As OpenAI and ChatGPT captured most of the headlines, Anthropic quietly but very openly was building a product for corporations and other business entities to use to extract more profits and cost savings or whatever. Accordingly, Anthropic was also the one major AI company that had no apparent interest in intruding into education systems, because there’s very little money to be made there, let me tell you.

And this strategy by Anthropic…appears to have been a wild success? Once considered the Lyft to OpenAI’s Uber, Anthropic has now vaulted ahead in earned revenue, prospective valuation, and maybe most importantly, the amorphous and ephemeral but nonetheless vitally important sense of being the “market leader.” AI as a consumer product has flailed around for three years, but at least for the moment, there’s every indication of a genuine “product-market fit”—business-speak for “shit someone or something will pay for”—around enterprise-facing AI agents, and of those, Claude is considered best of the bunch. And as I covered here, all the major AI companies appear to be pivoting away from chatbots toward some version of this “agentic” business model.

Given all this, why has Anthropic dropped Claude for Teachers on us now? It’s a consumer-facing play that makes no obvious sense to me. We’ll get to what exactly Claude for Teachers comprises momentarily, but just as a general matter, it seems odd timing to “lean in” on this particular AI use when the backlash to AI and edtech in schools continues to grow. This appears misaligned to Anthropic’s business model and a recipe for future PR headaches.

Or maybe not even future ones, as Claude for Teachers is already a privacy intrusion nightmare machine. In order to sign up, one must “get verified” as a teacher, which means Anthropic is asking educators to upload pictures of their staff IDs. Worse, as Mark Racine observed, Anthopic is also encouraging teachers to “upload student rosters, diagnostic data, and attendance.” Perhaps this hints as to why Anthropic is doing this—these companies are addicted to datasets, after all—but whoo-boy are they playing with fire with this blatant intrusion on student privacy rights.

Which is why I was heartened to see Randi Weingarten, president of the American Federation of Teachers, take a strong stand against Anthropic’s efforts to further infuse AI into education. Just kidding! From the news release: “We’ve been working with Anthropic on a Gold Standard that sets out industry best practices for safety and privacy in K-12 education,” said Weingarten. We’ll come back to this later.

But first, what about Claude for Teachers as a product—is it compelling? Reader, it is not. With help from one of my special secret-agent educator moles, I managed to gain access, and it appears Anthropic spent all of five minutes creating a teacher-facing “skin” for Claude. Here’s the landing page:

If you click on “plan a lesson,” you’ll then be taken to this:

That’s it, that’s the level of “customization” for teachers provided in Claude for Teachers, this single pre-baked prompt. Everything else beyond this is just regular ol’ Claude. Wheeeee.

Look closely, and you’ll notice Claude for Teachers also includes a few listed “Connectors,” included to something called Learning Commons. As best I can tell, this is the rebranded education division of the Chan Zuckerberg Initiative that now offers edtech companies something called a “Knowledge Graph” that purports to unify all 50 state academic standards here in the US. From the Learning Commons site:

This is bizarre for multiple reasons. First, no public school teacher needs this, there is zero value to them in having a “coherent, interconnected network” of unified academic standards—what matters is the standards in their particular state. But second, and far more bafflingly, surely both Anthropic and CZI are aware that one of the single biggest political controversies in education politics in the US was around the Obama Administration’s attempt to create a coherent, interconnected network of academic standards called the Common Core? And that the radical right-wing pushback to this effort, led by Glenn Beck and Michelle Malkin, was in some sense an early warning indicator for the rise of Trump, MAGA, Moms for Liberty and the like? Why would Anthropic and CZI want to go anywhere near this third rail again by trying to create what appears to be de facto national standards?

To further beta Claude for Teachers, I fed it one of my favorite test tasks, asking it to plan a lesson around the Civil Rights Movement. It provided what all these tools provide, a completely anodyne lesson structure that would be destined to go awry in the hands of all but the most expert of teachers. But I then prompted it to “provide guidance as to how I should teach this lesson given I am in red state and many of my students have parents who support Trump,” which resulted in this internal monologue on Claude’s part:

This is a crystal example of what AI scholars such as Abeba Birhane are warning about when they point out how AI is both encoding and shaping social prejudices at scale. To wall off “contemporary political issues” from our recent history is to make a choice, and here, Anthropic’s algorithm, programmed by Anthropic’s software engineers, is making it very clear that teachers should not connect the past to the present. Shudder.

One last thing on Claude for Teachers’ functionality: Anthropic purportedly is providing teachers with “free access to premium Claude capabilities,” but if you’re looking to make use of the fabled Fable, Anthropic’s most advanced coding agent, you’ve got all of three days from sign-up before access will be cut off. And even while Fable is briefly activated, the usage limits are severe—it timed out on the sole task I gave it after 30 minutes, and promptly cut off all token credits.¹ I mean, what is the point?

Oh and if you’re wondering, I didn’t sit through the online “AI fluency” course developed by Teach For America, sorry not sorry. I did, however, take a quick peek at the obligatory end-of-course quiz, and this item made me spit out my milk tea:

So what’s happening here is that to be AI fluent, according to Anthropic and Teach For America, individual teachers need to bear the responsibility for ensuring compliance with FERPA, even though the law itself applies to school districts and other legal entities (not individual educators). Or as Claude puts it:

Right, so Claude for Teachers is FERPA compliant if and only if teachers take on the work of anonymizing their student data each and every time they enter it into Claude. This is what Randi Weingarten is describing as the “Gold Standard” for privacy protection? What are we doing here?

It’s all very weird and annoying and I’d like to go back to New Mexico.

---

UPDATE 7/27/26: Alert CogRes reader Connie Ma has alerted me to this essay by Dr. Joe Phillips, a school district administrator (I think), on why the district where he works is banning Claude for Teachers. The underlying FERPA privacy issue he flags aligns with what I’ve highlighted here.

1. The task, you ask? Of course I asked Claude Fable to take on Zork! It scored 58 points, better than my trials with Codex, but it was also completely unclear how it was engaging with the game. AI agents are confusing.