Teachers Unions Training 400,000 Educators on AI While Striking for Protection Against It in 2026

On March 18, fifty teachers gathered at union headquarters in New York City to learn how to build autonomous AI agents. The same week, 35,000 Los Angeles teachers announced they'd strike on April 14 — with protections against AI replacement listed among their demands.

Both events were organized by teachers' unions.

This isn't a contradiction. It's a strategy. And it's redrawing the lines between workers and technology faster than any industry has managed before.

The $23 Million Bet

The American Federation of Teachers — 1.8 million members, second-largest teachers' union in the US — launched its National Academy for AI Instruction last summer with $23 million from three companies: Microsoft, OpenAI, and Anthropic. The goal: train 400,000 K-12 educators to use AI over five years.

Not the basics. Not "here's how to ask ChatGPT to write a quiz." The March 18 session in New York trained teachers to build agentic AI tools — autonomous software that can stress-test lesson plans for content gaps, brainstorm alternative interventions when a teaching approach fails, and track individual student progress across weeks.

Jing Liang Guan, a science teacher at Brooklyn Science and Engineering Academy, explained the difference to Education Week: he doesn't want a tool that writes his lessons. He wants one that helps him find the weak spots in lessons he's already written.

Lois Torres, a preschool paraeducator, wants an agent that can help her and her co-teacher brainstorm on the fly when a behavioral intervention isn't working. "Sometimes it takes a lot for the teacher and a para like me to figure out how to help a kid," she said. "And then, when something doesn't work, it's like, OK, now we've got to brainstorm on the fly."

The share of US teachers using AI-powered tools nearly doubled from 2024 to 2025. Six in ten now use AI in their practice, according to EdWeek's national survey. A RAND Corporation study found 53% of English, math, and science teachers used AI for school in 2025. But most still use it for surface-level tasks — generating lesson templates, organizing calendars, drafting parent emails.

The academy wants to push past that floor. And it's doing something unusual: it's teacher-led. Educators train other educators, with limited support from the tech companies bankrolling it.

The Strike Demand Nobody Expected

Three days after the New York AI training, United Teachers Los Angeles set an April 14 strike deadline. Their demands include pay raises, smaller classes, more mental health staff — and explicit protections against AI replacing educators.

UTLA isn't alone.

In St. Paul, Minnesota, the Saint Paul Federation of Educators became one of the first K-12 unions in the country to win AI-specific language in a contract. It started when the district's HR director couldn't promise that AI wouldn't lead to job cuts. Union president Leah VanDassor called the AFT's field programs director and got "pages and pages" of suggested contract language.

The result: binding provisions that insist human educators — not algorithms — stay at the center of teaching and learning.

In San Diego, the education association's new 2025-2027 contract includes Article 2 protections against educators being replaced by AI. In Canada, the Public Service Alliance pushed for 15 new AI-related clauses in federal contract negotiations, including one establishing that AI can't substitute for public-sector employees.

"We're not working in a factory where we're trying to shave seconds off the production line," VanDassor told the AFT. "We're making sure people come to school and are educated by other people."

The Paradox That Isn't

Take the money from AI companies. Train your members on AI. Then bargain contract language to prevent AI from replacing them. At first glance, it looks like teachers want it both ways.

Look closer and the logic is tight.

Rob Weil, the AFT's director of field programs and CEO of the National Academy, put it plainly: for years, school districts treated technology as a "management right" — administrators decide what to buy and that's it. That worked for laptops. AI is different. It touches instruction, assessment, student relationships, professional judgment.

"Everybody's guardrails always say we have to keep the human in charge," Weil said. "OK — bargain it. Put it in the contract."

The St. Paul contract illustrates the tension. The district adopted a phonics curriculum called UFLI with a built-in AI component. Teachers must feed spelling test results into the system, which then groups students and suggests instruction. One teacher's response: "That's my job. I know who these kids are."

The objection isn't that AI can't group students by spelling errors. It can. The objection is that making the system mandatory removes the teacher's professional judgment from the loop — and once it's gone, it's hard to argue the teacher is still essential.

Bloom's Ghost in the Machine

Underneath this tug-of-war sits a 42-year-old research finding that Silicon Valley can't stop talking about.

In 1984, educational psychologist Benjamin Bloom found that students receiving one-on-one tutoring performed two standard deviations better than classroom-taught peers. The average tutored student outperformed 98% of conventionally taught students. He called it the "two-sigma problem": we know tutoring works, but we can't afford to give every child a tutor.

AI companies pitch their tools as the solution. Khan Academy's Khanmigo — $44 per year — uses GPT-style models to guide students through problems rather than just giving answers. Teachers report students come to class more prepared. Duolingo's AI adapts to each learner's level in real time.

But the evidence is more modest than the pitch. Matthew Kraft at Brown University reviewed decades of educational interventions and found most produce effects of 0.1 standard deviations or less — a fraction of Bloom's two-sigma finding. No AI tutoring system has come close to matching human one-on-one instruction at scale. Not yet.

The gap between promise and proof is exactly where teachers' unions are staking their ground. Train us so we're in the driver's seat, they're saying. But don't replace us with tools that haven't earned it.

The 92% Problem

While teachers negotiate over AI, their students aren't waiting.

Global student AI usage jumped from 66% in 2024 to 92% in 2025, according to DemandSage. In the UK, 88% of students admitted to using generative AI for tests. A Frontiers in Education systematic review published this week found AI-enabled personalization improves both academic performance and student motivation — but also found the field lacks longitudinal studies on what happens when students lean on AI throughout their education.

The classroom is already an AI workplace whether teachers bargained for it or not. The question is who controls the terms.

Goldman Sachs called it last week: "The big story in 2026 in labor will be AI." Education is where that story is playing out first — not in tech companies or factories, but in rooms full of children and the adults trying to teach them.

Learn the Tool, Draw the Line

The AFT's dual strategy — embrace and constrain — might be the most sophisticated labor response to AI anywhere in the world. It's also almost entirely a US story. Canada's PSAC is pushing similar contract language, but no other country's education system is running simultaneous campaigns to train hundreds of thousands of teachers on AI while bargaining explicit protection against it.

There's something quietly radical about it. The teachers aren't saying AI is bad. They're not even saying it's dangerous. They're saying: this changes who has power in a classroom, and we intend to keep it.

Yasheema Cook, a special education teacher in New York City, is building AI agents to help manage individualized education programs for her 12th graders with autism and intellectual disabilities. She finds it a careful balancing act — giving the AI enough context on student needs to be useful while protecting their privacy. New York City public schools still haven't released their promised AI guidance.

The guidance gap is everywhere. RAND's data shows AI use in schools surging while formal policy lags far behind. Teachers are building the plane while flying it — and bargaining the safety manual at the same time.

Randi Weingarten, the AFT president, framed it as a race: "There is a real demand from educators to learn so that they are in the driver's seat for AI, as opposed to the companies or districts or the technology itself."

The April 14 LAUSD deadline will test whether that framing holds. If 35,000 teachers walk out and AI protections are on the list, it won't be a footnote. It'll be the first major strike in history where workers demanded the right to control how they use the technology that's supposed to make them obsolete.

That's not a contradiction. That's the future of work arriving in a classroom.