Anthropic Launched Claude for Teachers. We Blocked It for Our District.
We made that decision quickly and thoroughly.
Not because Claude is a bad AI model. Quite the opposite. I believe it’s one of the best AI models available today.
We blocked it because I believe Anthropic made one of the biggest K-12 go-to-market mistakes I’ve seen in years.
The frustrating part is that it was completely avoidable.
Here’s What I Saw
As I read through Anthropic’s launch materials, a few things immediately stood out.
Claude is described as:
Safe with student data.
Built with FERPA in mind.
Designed specifically for teachers.
Anthropic’s own examples encourage teachers to work with student assessment data, attendance information, classroom notes, and even reference IEP meeting notes.
Then, farther down the same page, Anthropic says:
“Claude for Teachers is for individual educators. A dedicated offering for schools and districts is coming soon.”
That’s where they lost me.
The district offering should have come first.
Not later.
If your product encourages workflows involving protected student information, districts need governance before teachers need access.
Not the other way around.
Teachers Were Put in the Middle
This is my biggest concern.
Teachers aren’t privacy attorneys.
They aren’t FERPA experts.
They’re teachers.
When a company tells teachers a product is safe with student data, built with FERPA in mind, and then demonstrates workflows involving student information, many educators will reasonably assume those workflows are appropriate.
In my opinion, Anthropic has put teachers in a position where they could inadvertently make unauthorized disclosures of protected student information while believing they were using the product exactly as intended.
I am not suggesting teachers would intentionally violate FERPA.
I’m saying they could reasonably believe they were doing exactly what the product was designed to do.
That is not a position teachers should ever be placed in.
The District Treats It Like a Data Breach
When an unapproved application receives district data, we do not start by asking how good the AI model is.
We handle it exactly like a data breach.
Because, from the district’s perspective, protected information may have been disclosed to an unauthorized third party.
We immediately need to know:
Which employees created accounts?
What information was uploaded?
Which students were affected?
Were IEPs, evaluations, behavior records, assessment data, or other protected records involved?
Can the data be permanently deleted?
Can the vendor prove it?
Do we have notification, legal, or regulatory obligations?
Whether a specific incident meets the legal definition of a reportable data breach depends on the facts and applicable law.
Operationally, we don’t have the luxury of debating terminology first.
We contain it.
We investigate it.
We determine the scope.
We document the response.
We notify the appropriate parties when required.
We demand deletion of district data.
The teacher clicked upload.
The district owns the investigation.
The district owns the legal review.
The district owns the communication with parents.
The district owns the regulatory response.
The district owns the cleanup.
The platform gains another user.
The district gains another incident to investigate.
Privacy Is More Than Model Training
Anthropic deserves credit for making meaningful privacy commitments, including that educator conversations are not used to train their models.
That’s the right decision.
But protecting student information is about much more than model training.
The first question isn’t:
“What happens after the data gets there?”
It’s:
“Should the data have been disclosed in the first place?”
Those are two completely different questions.
From a district perspective, student data isn’t considered protected simply because it won’t be used to train an AI model.
It also has to be disclosed under the appropriate legal, contractual, and governance framework.
Then I Got to the Verification Process
This one surprised me.
Depending on the verification pathway, educators may be asked to provide a school-issued ID, government-issued ID, employment verification, a contract, or a recent pay stub.
Think about that.
School districts already know who their employees are.
We already have HR systems.
Google Workspace.
Microsoft Entra.
Identity providers.
Clever.
ClassLink.
Why create another process that asks teachers to submit sensitive employment documentation to another third party?
Even if handled securely, I don’t understand why that additional risk needs to exist.
This EdTech GTM Playbook Is Getting Old
Unfortunately, this isn’t unique to Anthropic.
I’ve watched this same playbook repeat itself for years.
Launch free teacher accounts.
Drive rapid bottom-up adoption.
Get thousands of educators using the platform.
Then walk into the district office and say:
“Hundreds of your teachers are already using our platform. If you want visibility, governance, identity management, administrative controls, and district oversight, you should purchase the enterprise version.”
That isn’t responsible K-12 adoption.
It puts districts in a position where they have to react after adoption has already started instead of evaluating the platform before sensitive student information may be shared.
Governance becomes the upsell instead of the foundation.
That’s backwards.
Governance isn’t an enterprise feature.
It isn’t a premium add-on.
It is the prerequisite for adoption.
School districts are the customer.
More importantly, they’re the legal stewards of student data.
Teachers, students, and staff are the district’s users. They are not an independent customer base for vendors to acquire around the district.
This Didn’t Have to Happen
One thing that surprised me was the apparent lack of engagement with many of the district technology leaders I regularly work with before this launch.
When Google and OpenAI were preparing major K-12 initiatives, their GTM leaders reached out to me and many other district CIOs, CTOs, and K-12 leaders.
They wanted us to challenge the strategy.
They wanted us to find the holes.
They changed things because of those conversations.
I’m not saying Anthropic didn’t engage with anyone.
I’m saying this rollout doesn’t feel like it was pressure-tested by enough district leaders who live with student privacy, procurement, governance, and technology risk every day.
Because if it had been, I believe someone would have asked one simple question:
“Could a teacher reasonably and falsely believe they’re authorized to upload protected student information because of how we’ve presented this product?”
If the answer is yes…
The rollout wasn’t ready.
Claude Deserved Better
I don’t think Anthropic has a technology problem.
I think they have a K-12 strategy problem.
Claude is an exceptional AI model.
I genuinely want it to succeed in education.
But great AI doesn’t excuse a poor go-to-market strategy.
District office should not be an afterthought.
Governance should not be an upsell.
And teachers should never be placed in a position where they inadvertently create FERPA compliance issues while believing they’re using a product exactly as intended.
Claude deserved a better rollout.
I hope Anthropic fixes it.
I've seen this model before and am always leery of vendors who do this. I actually had the honor of having one vendor try this on Dr. Joe when we were working together. Needless to say, that vendor didn't make it to Procurement. There is one unique danger with Claude entering this space, though. Many people would assume that they are fundamentally altering their LLM for education. They are not. This is just another instance of wrapping some skills and tools into a core model and marketing it.
In our district, we set an extremely high bar for student-facing AI tools. If they don't offer real-time threat detection and escalation and direct conversation moderation, we're not using it. And we're not interested in farming out the detection to the gaggle of tools (pun intended) that operate in the student security space. If the AI tool doesn't do it, we won't use it. And for now, none of the major players out there (OpenAI, Anthropic, Google, Microsoft, etc.) are doing it.
...am I a nerd because I was excited every time you used the word "governance?" FERPA has always been the "gold standard" in this field; do you believe there is a way to provide powerful AI tools into the classroom while upholding iron-clad protections for students?