Can a $23M, vendor-funded AI academy help 400,000 teachers, or just help its funders?

In a windowless Manhattan conference room last summer, Queens high school English-as-a-new-language teacher Darius Saczuk sat down to learn the technology he had walked in calling the enemy. "It's killing our jobs," he said. He came anyway, and the sentence he wrote down in the training was a different one: "In order to retain my job, I need to know how to use AI because it's not going away." Saczuk had come to the first session of the American Federation of Teachers' new National Academy for AI Instruction, a year-old program that the union is now scaling into the first union-scale AI training effort for US teachers (The Guardian).

The academy is a $23 million program funded entirely by Microsoft, OpenAI, and Anthropic. The American Federation of Teachers, with roughly 1.8 million members, is one of the two largest US teachers' unions. Its stated goal is to train 400,000 educators over the next five years, the equivalent of about one in five American teachers, on both the uses and the limits of generative AI, and to give teachers a hand in deciding how the technology is used in their classrooms. AFT president Randi Weingarten has characterized teaching educators about AI as a "moral imperative."

The academy lands in a near-empty professional development shelf. A Pew Research Center survey of 1,458 US teens in the fall of 2025 found that about 54% of teens now use AI for schoolwork and that roughly six in ten think AI cheating is regular at their schools. An NPR/Ipsos poll of 545 K-12 teachers in 2026 found that only about four in ten say their schools offer any AI professional development at all, and that roughly half say their school has not offered any AI guidance or that they are unsure whether it has. Teachers' labor contracts, EducationWeek has reported, almost never mention AI.

The vendors have moved into that gap faster than the districts, and the AFT has chosen to work with them rather than wait.

The risk critics point to is not the training itself. It is the curriculum. When the funder of a professional development program is also the company whose products the program teaches, the program can become a definition war. "AI literacy," as written by the funder, tends to include using the funder's products correctly, the way "Microsoft Office proficiency" once meant fluency in Word, Excel, and PowerPoint. Audrey Watters, author of Teaching Machines and a longtime critic of education technology, has framed the AFT's arrangement as a transfer of political power to big tech and called it dangerous for educators.

Watters's capture argument is the strongest version of the critique. It is also falsifiable. The academy will prove it is serving teachers rather than its funders if, over the next two years, at least three things happen: the published curriculum is authored or reviewed outside the three vendors, not just branded by them; the academy publishes conflict-of-interest disclosures on every lesson, including which prompts, tools, and case studies the vendors have asked to add or remove; and AFT locals, not the national office, control how the training rolls out in their schools, including the right to refuse deployment of any vendor tool the bargaining unit does not approve.

If those three conditions hold, the Big Tech capture frame breaks. If they do not, the 400,000-educator figure is a marketing number attached to a vendor onboarding channel with a union logo, and the academy has taught 400,000 teachers the funder's version of the technology rather than their own.

The vendors have an obvious reason to be in the room. Microsoft, OpenAI, and Anthropic are competing to put their consumer- and enterprise-facing products into classrooms, and the first group of teachers to receive structured, union-branded AI training is a multi-year head start on the rest of the K-12 market. The AFT's bet is that the union's scale, its 1.8 million members, and its locals' control over contract language are enough leverage to make the vendor relationship a teacher training program rather than a sales channel. That bet has not yet been tested in public.

The closest comparable case is the longer-running relationship between textbook publishers and teacher associations, in which editorial independence and funder access are negotiated line by line in every edition. The AFT academy does not yet have a public record on that kind of editorial independence.

The next test is concrete. The academy's first full year of curriculum and PD delivery is already underway. The question is not whether the AFT can train 400,000 teachers. It probably can come close. The question is whether the curriculum those 400,000 teachers walk out with was written to teach them, or written to teach them the funder's version of the technology. By the end of the academy's first full academic year, the answer should be in the curriculum itself.