The Conflict-of-Interest Question in AI Training

Here is the arrangement, stated as plainly as it can be: several of the largest AI companies have put money into a national program, run through a major teachers' union, that trains educators to use AI tools — including the vendors' own. When you write it that sentence-first, the concern writes itself. The people who profit from adoption are helping fund the machinery of adoption. That is a textbook conflict of interest, and no amount of good intent makes it stop being one. The honest task isn't to wave the concern away. It's to weigh how much it should worry you, and against what.

Name the conflict without flinching

Start by refusing the two easy exits. The first easy exit is to say "everyone has conflicts, this is fine" — a shrug that treats a real structural problem as background noise. The second is to say "vendors funded it, therefore it's captured" — a reflex that treats funding as proof of corruption. Both skip the actual analysis. The interesting question lives between them.

The structure is what it is. When AI companies contribute to a union initiative that trains members on AI, the companies have an obvious interest in what that training says. Even with no one acting in bad faith, incentives bend curricula. A training program funded by tool vendors is under quiet pressure to present those tools as solutions rather than as trade-offs, to teach how to use them before asking whether to, and to normalize adoption as inevitable. Word In Black's reporting on the AFT's AI deal lays out exactly this worry, and it's not a fringe objection — it's the first thing a careful person should ask.

The conflict isn't that anyone is being bribed. It's that money shapes emphasis, and emphasis shapes what a generation of teachers comes to treat as normal. That happens without a single villain in the room.

There's a sharper version of the concern too. Teachers who feel skeptical of a vendor-funded rollout aren't paranoid; they're responding to a genuine misalignment of incentives — the same skepticism that surfaced when Oregon teachers called AI deals a "grift". Dismissing that skepticism as anti-technology misses that it's often just accurate reading of who benefits.

Now weigh the safeguards fairly

Having named the conflict at full strength, honesty requires giving the other side the same courtesy — because the arrangement has a design feature that materially changes the analysis: educators, not vendors, build the curriculum.

That distinction matters more than it first appears. A conflict of interest is dangerous in proportion to how much control the conflicted party has over the output. If AI companies wrote the training materials, the concern would be close to fatal — you'd be letting the seller write the buyer's guide. But a program where the money comes from vendors and the content is designed by working teachers inverts the most dangerous part of the structure. The people with classroom judgment, professional skepticism, and a union's institutional incentive to protect members are the ones deciding what gets taught. Vendors can fund the room; they don't get to script it.

Teacher-designed curriculum has a specific virtue here: teachers have every reason to teach the limits of these tools, because teachers are the ones who get burned when a tool fails in front of thirty kids. A vendor's training would gloss over hallucination, bias, and privacy landmines. A teacher's training has to cover them, because the teacher's credibility with the class depends on it. The incentive that funds the program and the incentive that designs it point in different directions — and the design incentive is the one that touches the actual lesson content.

It's also worth being precise about what's being trained. The skills a good program teaches — when to trust an AI output, how to check it, when to point the tool at your own prep versus a student's graded work — are portable across vendors. Learning to critically use GPT-5.1 makes you better at critically using Claude or Gemini too. That portability blunts the capture concern, because the training isn't teaching allegiance to one product; it's teaching a transferable literacy. The distinction between tool loyalty and tool fluency is the whole ballgame, and it maps onto the same teacher-versus-student divide explored in free for teachers, banned for students.

Where the residual risk actually lives

So the safeguard is real and it's load-bearing. But "the curriculum is educator-designed" is a claim that has to keep being true, not a one-time absolution. The residual risk isn't in the headline structure; it's in the slow places.

It lives in default settings — when a training program uses one vendor's product as the running example throughout, students of that program come away fluent in one interface and reaching for it by reflex. It lives in what gets omitted: a curriculum can be technically balanced and still never seriously entertain the option of "don't use AI for this," because that option is invisible to a program whose entire premise is AI adoption. And it lives in the quiet gravity of free — when OpenAI's free-through-2027 teacher offer removes the last friction, "which tool should we use" collapses into "the one that costs nothing," and the training's careful neutrality gets overwhelmed by economics no curriculum controls.

Those are real risks. But notice that none of them are solved by refusing the training, and most are made worse by it. A teacher who gets no structured, skeptical instruction doesn't become AI-free; they become AI-fluent by trial and error, learning from vendor marketing and colleagues' habits instead of from a curriculum designed to surface the trade-offs. The counterfactual to imperfect, vendor-adjacent training isn't pristine independence. It's untrained adoption, which is worse on every axis the conflict-of-interest critique claims to care about.

The verdict is a "yes, and"

So where does an honest reckoning land? Not on exoneration and not on alarm. The conflict of interest is real — vendors funding the training that drives adoption of their products is exactly the kind of entanglement that should make you check the fine print. Anyone who tells you it's nothing is selling something. And the safeguard is also real — educator-designed curriculum moves control of the actual content to the party with the strongest incentive to teach the tools' limits. Anyone who tells you the safeguard is theater hasn't looked at who writes the lessons.

The mature position holds both. Watch the conflict; it doesn't go away because there's a safeguard. Credit the safeguard; it isn't cosmetic. And judge the program by the thing that actually determines the outcome — not by who signed the check, but by whether the curriculum keeps teaching teachers to be skeptical customers rather than loyal ones. That's a question you answer by reading the materials, not the funding disclosure. The disclosure tells you where to look. It doesn't tell you what you'll find. And the difference between those two vendors' tools — what you'd actually be trained toward — is worth understanding directly, which is where the head-to-head comparisons come in.

Part 72 of 100 in the ChatGPT for Teachers series. Previously: Free for Teachers, Banned for Students: The Split. Next: Claude for Education: The Other Free Offer. Browse more builder insights or explore AI skills for education at aiskill.market.