Why a Union Took AI Vendor Money

There's a question sitting under the National Academy for AI Instruction that no amount of good intention makes disappear: why would a teachers' union take $23 million from OpenAI, Microsoft, and Anthropic to train its members on the very tools those companies sell? A union exists to represent workers, sometimes against the firms reshaping their jobs. Accepting money from those firms to accelerate the reshaping is, at minimum, a move that demands an explanation — and the American Federation of Teachers has given one. It's worth taking that explanation seriously, and then taking the skepticism seriously too, because both are more substantive than the caricatures suggest.

This is the uncomfortable pivot of the whole ChatGPT-for-Teachers story. The free tool is easy to evaluate. A union's decision to partner with the industry it might otherwise be checking is not.

The AFT's actual argument

The union's rationale is not "AI is great, let's help sell it." It's a bet about agency, and it goes like this: AI is arriving in classrooms whether teachers want it or not. The only real choice is whether educators shape how it arrives or have it imposed on them by administrators, ed-tech procurement, and students who are already using it. Given that framing, sitting out looks less like principled resistance and more like forfeiting the design decisions to everyone else.

So the AFT's stated position is "teachers in the driver's seat": members help design the curriculum, test the tools, and draft the guardrails, rather than receiving a vendor's finished training. The union frames the money as funding for educator-led development — the resources to build teacher agency into AI adoption while the norms are still soft. In this reading, refusing the money wouldn't protect teachers; it would just leave the field to be shaped by parties with no obligation to them.

The AFT's bet is that the way to protect teachers from AI is to put them in charge of teaching it — and that taking the vendors' money is the price of a seat at the table where the defaults get set.

There's a real logic here. Unions negotiate with employers they're structurally opposed to all the time; proximity isn't the same as capture. And the privacy work that accompanied the deal — vendors committing to FERPA-grade handling and no-training-by-default — is arguably the kind of concession you extract precisely because you're at the table, not despite it. Randi Weingarten's AFT has consistently argued that engagement produces better guardrails than boycott.

The skepticism it drew

The critics' case is equally coherent, and it starts with a structural observation rather than an accusation of bad faith. When a union takes money from vendors and then trains its members to adopt those vendors' products, the union has acquired an interest in the adoption succeeding. That interest may never override its duty to members — but it now exists, and it points the same direction as the companies' interests. That's the textbook shape of a conflict, and naming it isn't cynicism; it's the reason disclosure rules exist.

The sharper attacks go further. A rank-and-file caucus within an Oregon affiliate of the National Education Association — the AFT's larger rival union — has publicly opposed what it labels "AI grift," arguing that unions shouldn't be in the business of onboarding members to Silicon Valley's products at all. And outlets covering the deal, including Word in Black, raised pointed questions about a labor organization accepting funding from the same firms whose tools it's now promoting to a captive professional audience.

Two honesty notes are required here. First, the "grift" and conflict-of-interest framings are arguments made by critics, not settled findings — some of the loudest come from advocacy outlets like the Capital Research Center, which approaches teachers' unions from a distinctly critical political vantage. Treat them as attributed opinion to weigh, not established fact. Second, the criticism comes partly from a rival union, which has its own reasons to attack the AFT's highest-profile initiative. That doesn't make the critique wrong. It does mean it should be read with the same skepticism it directs at the AFT.

What actually determines the answer

Strip away the rhetoric on both sides and the question becomes empirical: does the "driver's seat" hold, or is it a comfortable seat in a car someone else is steering? The framing is only as good as the mechanism behind it, and there are concrete tests.

Who controls the curriculum? If educators genuinely author the materials and can teach transferable, tool-agnostic judgment — how to evaluate any AI tool, protect student data, and catch confident errors — then the driver's-seat claim has substance. If the syllabus quietly centers the funders' specific products because their in-kind engineers and compute shaped it, the critics have their evidence. The standardization dynamics that make this a high-stakes call are examined in 400,000 teachers, one AI curriculum.

Can the union still criticize the vendors? The cleanest disproof of capture would be the AFT publicly pushing back on OpenAI, Microsoft, or Anthropic when their products harm teachers or students. A partnership that permits real criticism is a partnership; one that doesn't is a sponsorship. That test plays out over years, not at launch.

Does it change what teachers can do? If trained educators end up better at protecting students and more skeptical of AI's failure modes, the money bought something worth having. If they mostly end up fluent in three companies' interfaces, the critics were describing it accurately.

The honest verdict

The fair conclusion resists both the union's press release and the critics' worst-case. Taking the money is a genuine gamble, not an obvious sellout and not an obvious masterstroke. The AFT is wagering that presence beats absence — that shaping AI's classroom debut from the inside protects teachers better than principled distance would. The critics are warning that presence can curdle into promotion, and that a union with a financial stake in adoption is a compromised guardrail.

Both can be true at different times, which is exactly why the mechanisms matter more than the mission statements. "Teachers in the driver's seat" is a testable claim, not a settled fact — and the honest position, a year in, is to hold it as a bet still being adjudicated. What happens inside the training rooms is where the answer gets written, which is why it's worth looking closely at the NYC hub where the teaching actually happens.

Part 59 of 100 in the ChatGPT for Teachers series. Previously: 400,000 Teachers, One AI Curriculum. Next: The NYC Hub Training America's Teachers on AI. Browse more builder insights or explore AI skills for education at aiskill.market.