5 Questions to Ask Any AI Vendor Before You Sign

Q: Why does it matter who funds a vendor's safety claims?

Because vendor-funded \"independent\" coalitions or safety credentials can misrepresent their actual objectivity, as shown by reporting on OpenAI's undisclosed funding of a child-safety coalition in 2026.

Drawing directly from the criteria NYC, LA Unified, and Chicago's school board candidates have applied to AI in 2026, five questions separate a defensible AI purchase from an exposed one: does it touch students directly, what evidence does it actually use, who funds any safety claims behind it, what happens when it's removed, and can you answer a parent's question about it in one clear sentence.

Five Questions 2026's AI Policy Wave Says You Should Ask Before Signing Any AI Vendor

Every major AI-in-schools controversy this year - NYC's moratorium, LA Unified's reversal, the Chicago pledge, the OpenAI coalition story, the Lower Merion lawsuit - has effectively written a due-diligence checklist for anyone about to sign an AI vendor contract. Here it is, made explicit.

1. Does this tool interact with students directly?

This is the single question that determines which regulatory and political category a product falls into. NYC and LA Unified both drew their restriction lines exactly here: student-facing generative AI is restricted; teacher-facing AI is not. A product with a chatbot interface a student logs into is subject to every safety, consent, and review requirement these policies establish. A product that only works on evidence a teacher already has - assessments, not conversations - isn't touched by any of it, by design, not by exemption.

2. What specific evidence does it actually use?

Vague answers here ("it uses AI to personalize learning") should be a red flag. NYC's own tool-review criteria name evidence of learning design and instructional impact explicitly. A defensible answer names the actual input: "It analyzes results from the assessments your teachers already give, and identifies the specific misconception behind each wrong answer." A vendor who can't name their actual input data source in one sentence likely can't defend it under review either.

3. Who funds the safety claims backing this vendor?

The OpenAI-funded "kids safety coalition" story is the case study here: a vendor's cited safety credentials, coalition endorsements, or "expert-backed" language are only as trustworthy as their funding source. Ask directly who funds any coalition, guidebook, or certification a vendor cites, and treat undisclosed or vendor-funded endorsements with the same skepticism the nonprofits in that story eventually applied.

4. What happens to student performance when the tool is removed?

Stanford's Generative AI for Education Hub has documented a "performance cliff" pattern: students perform well with certain AI tools and drop sharply once the tool is taken away, meaning the tool was doing the work rather than building the skill. Ask any vendor directly what evidence they have of student performance without the tool, not just with it.

5. Can you explain this tool to a parent in one honest sentence?

Every 2026 controversy - NYC's board meetings, Lower Merion's lawsuit, the general parent frustration Axios documented on September 7 - traces back to schools unable to give parents a clear, specific answer about what a technology does. Before signing anything, write the one-sentence answer you'd give a skeptical parent who asks "what exactly does this do with my child." If that sentence is vague, defensive, or requires a follow-up meeting to explain, that's information about the product, not just about your communication skills.

Why these five, together, matter more than any single one

A vendor can pass one or two of these questions and still be exposed. A tool that doesn't touch students directly (question 1) but can't name its actual evidence source (question 2) is still a black box a board can reasonably reject. A tool with clean data sourcing but vendor-funded safety credentials (question 3) still carries the OpenAI-coalition risk. The five together are closer to a complete picture than any one alone - which is exactly why 2026's various controversies each surfaced a different one of them, rather than all pointing at the same failure.

See Teacher-Facing AI in Action.

FAQ

What's the most important question to ask an AI vendor for schools?

Whether the tool interacts with students directly or only works on evidence teachers already have - this single distinction determines which regulatory and political scrutiny category the product falls into, based on how NYC and LA Unified drew their 2026 restrictions.

Why does it matter who funds a vendor's safety claims?

Because vendor-funded "independent" coalitions or safety credentials can misrepresent their actual objectivity, as shown by reporting on OpenAI's undisclosed funding of a child-safety coalition in 2026.

What is the "performance cliff" and why does it matter for vendor evaluation?

A pattern, documented by Stanford researchers, where students perform well using an AI tool but drop sharply once it's removed - indicating the tool did the work rather than building the underlying skill.

How can a school quickly assess if an AI tool is defensible to parents?

Try writing a one-sentence, honest explanation of what the tool does with a child's data and attention. If that sentence is vague or requires extensive follow-up, treat that as a signal about the product itself.