Natasha Singer has covered technology in schools for the New York Times for over a decade, and this week she published an essay on AI in schools. It opens at a Microsoft AI workshop for teachers, held at the United Federation of Teachers headquarters, taught partly through a cartoon featuring a cheerful Minecraft duck. Watching it, Singer recognized something from her years covering health care: the pharmaceutical seminars that companies ran for doctors to move brand-name drugs. She quotes a Georgetown professor who has studied that influence for decades: “Industry funds what helps industry.”
Her reporting lays out how deep the parallel runs. Tech companies fund the surveys that make their products feel urgent, sponsor the curricula, sit on the standards committees, and fund the parent and teacher organizations. There is no Raytheon AP Physics and no Pfizer AP Biology, Singer notes, but there is an Apple-supported AP Computer Science course built on Apple's programming language, and a Microsoft-backed version taught partly through Minecraft.
Three days after Singer's essay ran, Bill Gates published about 6,000 words arguing that AI needs limits it doesn't currently have, and that leaders aren't confronting the challenges adequately. “There is no plan to ease the entry into the AI era,” he wrote. Think about this: Reporter Natasha Singer, Microsoft co-founder Bill Gates, and Microsoft's president are all saying versions of the same thing in the same week.
The call is coming from inside of the house… the insiders are telling you the industry can't regulate itself.
So, is the answer to throw the companies out? Probably not, and honestly, it isn’t even realistic anymore. AI now ships by default on the devices in your classrooms. But if the people who built this industry say no plan exists, then you must write one that protects your students. This issue is about the two places to write it: an audit of who shapes your teachers' training, and a layer between the model makers and your students.
Some folks have reached out to us asking how to subscribe themselves or their colleagues to this newsletter. Click here to do so. We send a newsletter every two weeks about the latest happenings with K-12 and AI.
IN THIS ISSUE:
Big Tech has become one of the largest providers of teacher AI training, and almost nobody tracks the funding
The case for an independent layer between the AI companies and your classrooms
6,000
Approximate words in Bill Gates's August essay arguing AI needs limits the industry has not (and will not) set for itself. Source
$23 M
Funding for the American Federation of Teachers' National Academy for AI Instruction, provided by Microsoft, OpenAI, and Anthropic. Source
1 in 10
Share of US teachers the academy aims to train by 2030. Source
18%
Teachers who say they have received formal guidance on AI from their own administrators. Source
16
States in the Southern Regional Education Board, which now publishes independent AI procurement and evaluation guidance for districts. Source
Start with the scale. The AFT academy alone, funded with $23 million from Microsoft, OpenAI, and Anthropic, aims to train one in ten American teachers by 2030. Google, Microsoft, and OpenAI each run their own free educator training programs on top of that. Whatever you think of the arrangement, the math is plain: the companies selling AI to schools are on track to be the largest providers of AI training that your teachers receive.
Two things are true at once here, and Singer's reporting, to its credit, calls out both. The first is that this training fills a vacuum that’s open: only 18 percent of teachers report formal AI guidance from their own administrators, and Microsoft's own president acknowledged to Singer that industry filled the gap out of necessity and that it is not how public education should work. AFT’s president made a similar point in defending the academy: somebody has to train teachers, and somebody with money showed up. The second truth is the one medicine learned the hard way: training funded by the seller tends to teach the buyer to buy. Doctors' pharma seminars felt like education too. Medicine's answer was not to ban industry from the room. It was disclosure requirements and independent accreditation for continuing education, so a doctor always knows who paid for what they're learning. We have no equivalent, and this month, the national paper of record argues that we need one in K12.
OUR TWO CENTS
You can't audit the whole industry, but you can definitely audit your own district. Pull the list of every AI training session your staff received in the past year, and add a column: who paid for it. Vendor-led, vendor-funded through a nonprofit, union academy, university, or district-built. No judgment yet, just observation. Most leaders I talk with have never seen that list, which means the district's training strategy is whatever outside parties decided it would be.
Then set three rules going forward. Disclosure: every PD session states who funded it, the same way research papers disclose their funders. Balance: for every hour of product training, consider requiring an hour of tool-agnostic instruction, the pedagogy that survives a vendor switch. And ownership: the district, not the vendor calendar, decides the training plan for the year. None of this bans a Google or Microsoft workshop; some of them are useful. It makes them one ingredient in a diet your district planned, rather than the whole menu, chosen by the cooks with something to sell. And yes, apply every bit of it to us too.
—Tracy
Build the provenance list: every AI training your staff took last year, and who funded each one.
Add a funding disclosure line to every PD session, internal or external.
Consider a balance rule: tool-agnostic pedagogy hours alongside any product training hours.
Put PD strategy on your calendar as a district decision, made annually, before the vendor invitations arrive.
Understanding who’s paying for your training is one piece of the puzzle. The other piece of the puzzle is how AI itself reaches your students; here structure matters more than any policy. Right now, most districts consume AI in one of three ways: built into search, bundled with some other software, or offered through “free” accounts. The company that built the model is the only one who decides how it behaves with your students, what data flows back, and what the defaults are. That is a lot of trust to place in an industry whose incentives the NY Times article documented and whose own co-founder just wrote that nobody, his industry included, has a plan. It’s the Wild West.
The alternative already exists here and there, and it's growing. Chicago Public Schools began blocking unapproved third-party AI from its network in December, routing use exclusively through providers the district has vetted. New York City now bans AI for students below the 8th grade and requires AI vendors to disclose capabilities and bars student data from training their models. The Southern Regional Education Board, a 16-state compact, publishes independent procurement and evaluation guidance, with one commission leader advising that adults test every AI tool before students touch it. And at Boston University, a project called EVAL is building independent efficacy trials for education AI, with the stated goal of an "FDA for education." The pattern across all of it: someone whose paycheck doesn't come from the model maker deciding how the model reaches kids.
OUR TWO CENTS
I don't think districts should consume frontier AI raw. The companies building the most powerful models are in an AI arms race, and their consumer defaults reflect it. Between those models and a classroom there should be a layer that answers to the district: applying the guardrails, enforcing the data rules, and optimizing for learning instead of engagement. Sometimes that layer is a district running its own gateway, like Chicago. Sometimes it's a consortium, a nonprofit, or a company like mine.
But “independent layer” is about to become a marketing label the way “designed for education” already is, so here is the test I'd put in every RFP. Can you “fire” a model provider and switch, or are you wed to a single company? Do you really run it, with your rules controlling how it works, according to your policy? Does anything flow back to train the underlying models? And do the defaults favor pedagogy over engagement, hints over answers? A thin wrapper on one company's API, with that company's incentives intact, is just Big Tech with a markup, and some products calling themselves education layers are exactly that. Judge every company in this category, ClassCloud included, by that test, and be willing to fire the ones that fail it.
Your students need a layer of protection from the big tech companies instead of being the product of the big tech companies. Remember: when a product is “free”, you’re the product.
—Russ
Add the control test to your RFPs: model-agnostic, district-controlled contract, no data back to model training, pedagogy-first defaults.
Consider routing staff and student AI through district-controlled instances, so your rules travel with every conversation.
Borrow the SREB and Pennsylvania procurement questions rather than writing your own from scratch.
Teach the independence habit to students too. One Oakland teacher has students interrogate any technology with four questions: who built this, who profits, who benefits, and what happens in ten years.
Coding Kids: Big Tech's Battle to Remake Public Schools by Natasha Singer
The essay this issue opens with was adapted from this book, and was released September 8 from W.W. Norton. Singer spent over a decade covering tech-industry influence in schools, reporting that helped prompt California's landmark student data privacy law, and the book traces the whole arc: how coding education scaled through industry backing, and how the same playbook is now being run for AI. Publishers Weekly calls it a searing debut. You will not agree with every conclusion, and that's fine; the reporting underneath is the most complete map anyone has drawn of how products reach your classrooms. Pre-order it now and it arrives in time for your first cabinet meeting of the year.
Chat with ClassCloud
We’re listening. Let’s Talk!
This newsletter works best when it’s a conversation, not a broadcast. If you want to talk through how any of this applies to your district specifically—or if you have feedback on what would make this more helpful—just hit reply. We read and respond to everything.
Schedule a Virtual Meeting
Thanks for reading,
Russ Davis, Founder & CEO, ClassCloud ([email protected])
Tracy Walters, Director of Customer Growth ([email protected])
ClassCloud is an AI company, so naturally, we use AI to polish our content.
