Two pieces of AI-in-schools news crossed my desk this week, and they belong together.

On November 10, the AFT and UFT announced a legally binding AI safety and privacy agreement with Anthropic covering districts that use Claude for Teachers. A week earlier, Google announced it is expanding its free Google AI Educator Series to more countries.

Both are good news. Stronger data protections are overdue, and free AI training for teachers lowers the barrier to getting started. I’m not here to knock either one.

But read the fine print on the Anthropic deal and you’ll notice something. The hardest part gets handed back to you.

Key takeaways

  • The AFT/UFT agreement with Anthropic makes student data protections and “educators stay in charge” principles enforceable for districts using Claude for Teachers.
  • According to EdWeek, the agreement also calls on districts to set policies and train staff so humans review AI-informed decisions about grading, discipline and class placement. That’s a district job, not a vendor job.
  • Free, vendor-built AI training for teachers is a useful on-ramp, but it teaches one company’s tools. It can’t build your school’s judgment, policy or culture.
  • TNTP’s research found districts spending heavily on teacher development without consistent improvement. More PD hours aren’t the goal. A clear picture of what better looks like is.

What the vendor agreements actually ask of schools

Here’s what the AFT says the agreement guarantees: student data is used only for authorized educational purposes, it’s never sold or used for targeted advertising, schools control retention and deletion, and companies must explain in plain language how their products work and where they fall short. It builds on a similar standard Microsoft signed in September, and the union says it’s still negotiating with OpenAI.

Now the part I want every leadership team to read twice. EdWeek’s Sarah D. Sparks reports that the agreement calls on districts to set policies and train staff so that a human reviews AI-informed decisions on grading, discipline or class placement. As EdWeek points out, that protection depends on educators being trained, not on the tools.

Think about what that means in practice. A contract can say “educators stay in charge.” Only your people can make that true on a Tuesday afternoon when a teacher is staring at an AI-suggested grade and 140 essays still to read.

Who in your building knows what a good human review of an AI-assisted grade looks like? Who decided? Who taught it?

That’s not a privacy question anymore. It’s a professional learning question.

Free AI training for teachers is an on-ramp, not a road

I’ll admit it: if I were a principal with no PD budget left in November, a free, on-demand course from Google would look pretty good. And it might be exactly the right first step for some of your staff. Google says its India sessions alone trained 80,000 educators in September. That’s real reach.

But let’s be honest about what any vendor’s training is designed to do. It teaches you to use that vendor’s tools well. That’s a legitimate goal. It’s just not the same goal as yours.

Your goal is a staff that can make good decisions about AI no matter whose logo is on the screen. A staff that knows when to use it, when not to, how to check it, and how to explain those choices to students and parents. Google is migrating Gemini Gems to Skills starting November 17, so anyone trained to build Gems last year is relearning this month. Training tied to a feature has a short shelf life. Judgment doesn’t.

And notice who isn’t in the room for a self-paced teacher course: your leadership team, your board, your IT director, your counselors, your parents. The agreement’s human-review expectation lands on all of them.

What the research says about “more PD”

Before anyone responds to all this by adding three more AI sessions to the calendar, it’s worth remembering TNTP’s 2015 study The Mirage. Studying large districts, TNTP estimated they spent an average of nearly $18,000 per teacher, per year on development, with teachers reporting roughly 19 full school days a year in development activities.

The finding that stung: most teachers didn’t appear to improve substantially from year to year, and no type, amount or combination of development activities, including the job-embedded kind, reliably separated the teachers who improved from those who didn’t.

I don’t read that as “PD doesn’t work.” I read it as a warning that volume isn’t a strategy. Hours, sessions and sign-in sheets aren’t evidence of anything. If you can’t describe what better AI practice looks like in your classrooms, and you’re not checking whether you’re getting there, more training just adds to the pile.

The same honesty applies to my own numbers. In the end-of-workshop feedback I’ve collected from 118 educators since 2023, 97% said they were already planning to implement, or very likely to implement, what they learned. I’m proud of that. But it measures how people felt on the way out the door, not what changed in March. Closing that gap is the whole job.

What school leaders can do this month

  1. Pull your AI contracts and read them. Find out which AI tools your district has agreements with, and whether the new Microsoft or Anthropic standards apply to them. EdWeek notes that tools built on top of these models by other companies aren’t covered.
  2. Name your “consequential decisions.” List where AI could touch grading, discipline, placement or special education in your system. Those are where human review needs to be defined first.
  3. Write down what a good human review looks like. One page. What does a teacher check before accepting an AI-assisted grade or comment? Then ask five teachers if it’s realistic.
  4. Audit your AI PD for vendor dependence. How much of what your staff has learned is tied to one product’s features? What would survive if that product changed next semester?
  5. Pick one measure that isn’t attendance. Decide how you’ll know AI practice is improving, whether that’s classroom walkthroughs, student work samples or a short teacher survey repeated each term.
  6. Put the board and parents on the calendar. If humans are responsible for AI-informed decisions, the people who govern and trust your school need to understand how that works.

If you want a framework to organize this, our K12 AI implementation guide walks through leadership, governance, teacher workflows and student AI literacy in order. And if you’re mid-purchase, revisit the five questions to ask before you buy another AI tool.

Where an outside partner helps

I’ve been doing long-term partnerships with schools since 2009, starting with 1:1 laptop rollouts. Here’s what that wave taught me: the devices didn’t change learning. Sustained support for teachers did. The companies selling the laptops offered training then, too, and it was useful for exactly what it covered.

What a partnership adds is the part no vendor and no single course can own: continuity across years, an outside view drawn from schools in many countries, and no product to sell you. Every contract I do is different, which is why I call it a partnership. Across a year it might mean working with your leadership team on what “humans stay in charge” means in your policies, coaching teachers through real AI-assisted grading and feedback, a board session, a parent night, and student sessions on AI literacy. Face-to-face, virtual, or a mix, depending on the outcomes you’re after. The aim is to build your internal capacity, not replace it.

That’s how the Alaska Staff Development Network work grew: a five-part AI series in Fall 2024, run again in Spring 2025, with advanced sessions added on reimagining assessment and learning with bots. One participant summed up where good training should land: “…AI is doable, manageable and an ally- if we know how to wisely use it.” That “if” is the work.

Tricia Friedman and I dug into the cost of standing still on the Shifting Schools podcast in Crossing the Line of Falling Behind: Where Is Your School on the AI Curve? And on Human Still Required, I wrote about why a district’s AI strategy can’t depend on a vendor’s off switch. The same logic applies to training.

FAQ

Is free vendor AI training for teachers worth using?

Yes, as a starting point. It’s a low-cost way for teachers to build basic skills with a specific tool. It doesn’t replace school-specific work on policy, human review of AI-informed decisions, or AI literacy for students, leaders and families.

What does the AFT agreement with Anthropic require of school districts?

According to EdWeek, it calls on districts to set policies and train staff so humans review AI-informed decisions about grading, discipline or class placement. The company commits to data and transparency protections; the human-review piece depends on the district.

Why hire an AI consultant for schools instead of relying on vendor training?

An independent partner works across tools, brings patterns from many schools, and stays with you as platforms change. The goal is building your staff’s judgment and internal capacity over time, not teaching one product.

The bottom line

The contracts are getting better. The free courses are getting bigger. Good.

But a contract can’t train your teachers, and a course can’t write your policy. “Humans stay in charge” is a promise only your people can keep.

So here’s my question for your leadership team: if someone asked you tomorrow who’s responsible for checking an AI-informed grade in your school, and how they learned to do it, what would you say?

If that answer isn’t clear yet, book a 30-minute call with me. Pricing and packaging are tailored to each school, so reach out and we’ll talk through what a partnership could look like for yours. You can also email me at jeff@jeffutecht.com.

Sources

  • AFT, “AFT, UFT and Anthropic Sign Binding AI Safety and Privacy Agreement for Schools,” November 10, 2026. aft.org
  • Sarah D. Sparks, “Anthropic Agrees to Protections for Teachers Who Use Claude,” Education Week, November 10, 2026. edweek.org
  • Jennie Magiera, “Expanding AI educator training to new countries,” Google Blog, November 3, 2026. blog.google
  • Abner Li, “Gemini app replacing Gems with skills in November,” 9to5Google, September 27, 2026. 9to5google.com
  • TNTP, “The Mirage: Confronting the Hard Truth About Our Quest for Teacher Development,” August 4, 2015. tntp.org (executive summary PDF)