By Sagar Shankaran, Founder of CallSphere
Admissions, placement tests and proctoring put private schools inside 2026 AI rules. The inventory, the file, the parent notice, and what is out of scope.
Key takeaways
You run a 340-student independent day school outside Columbus. You have never sold anything to anyone in Europe and Brussels is not your regulator. You are ready to skip this one. Before you do, check whether any of these three sentences is true of your school. At least one is true of most independent schools with an international line on the enrollment report.
An applicant filled out the inquiry form on Ravenna from an apartment in Madrid. A family posted to Frankfurt is applying for a Grade 7 seat, and their son sat your placement test at his kitchen table while a proctoring tool watched him. Your July exchange partner outside Lyon sends fourteen students and you send back a written evaluation of each one, which their school puts in a file.
The EU AI Act's high-risk and transparency obligations carry a compliance date of 2 August 2026, and they reach organisations outside Europe whose systems affect people inside it. Education is not a footnote in that law. It is one of the named high-risk categories, alongside credit and hiring.
In plain English: if a computer system helps decide who gets a seat at your school, who gets financial aid, or what grade a student receives, 2026 law treats it as a high-stakes system — and the school using it, not the company that sold it, carries most of the paperwork.
The first is admissions. If any part of your reading process sorts, scores, ranks or flags applicants — a tool that summarises a file, a system that predicts likelihood to enroll, a screener that reads recommendation letters — and one of those applicants is sitting in the EU when the decision lands, the rules reach that decision. Your campus being in Ohio does not matter.
The second is testing. Remote proctoring on an entrance exam, an automated scorer on a writing sample, an adaptive placement test that decides which math section a child enters — all of it sits in the high-risk category covering admission, learning outcomes and placement level. If the child takes it from Europe, you are the one using the tool.
The third is the one nobody thinks about: anything that reads a student's face or voice to guess how they feel — attention-monitoring in a classroom, "engagement scoring" on a remote lesson, mood detection in a counseling tool. Emotion-guessing inside a school is not a high-risk category to be documented; it is on the prohibited list, and that ban landed well before this summer. If a vendor is pitching it, the answer is no for every student, European or not.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent in your browser — 60 seconds, no signup.
The distinction that matters is whether the tool touches a decision about a person. Grammar help on a newsletter is not a decision. A tour-scheduling agent is not a decision. A tool that ranks 180 applications for 46 seats is, and so is one that helps the Financial Aid Committee sort who gets the last $60,000 of the aid budget.
That one test — does it touch admission, aid, grading, placement, discipline or hiring — sorts ninety percent of what runs in your building in an afternoon. The rest needs judgment, which is where your attorney earns the fee.
Give this to the Registrar, not the Director of Technology alone. The Registrar knows which systems hold student records; the tech director knows which are paid for. Get both lists in one room before reading season.
flowchart TD
Z["List every tool: Ravenna, Blackbaud, FACTS, proctoring, website chat"] --> A["Registrar names an owner for each one"]
A --> B{"Does it help decide admission, aid, placement or a grade?"}
B -->|No| C["Low-risk log: date added, owner, one line on what it does"]
B -->|Yes| D{"Could the person affected be sitting in the EU?"}
D -->|No| E["State-law file: written notice, named human reviewer, decision record"]
D -->|Yes| F["High-risk file: vendor instructions, named reviewer, saved activity logs"]
E --> G["Head of School signs the inventory before reading season opens"]
F --> G
C --> G
The output is one spreadsheet, six columns: tool, owner, what decision it touches, whether a European family could be on the other end, the named reviewer, where the records live. That is what you hand to counsel and to your accreditation visiting team.
The European deadline gets the attention, but the statutes that took effect at home on 1 January 2026 will touch you first. Texas TRAIGA and California SB 53 both went live that day, and Colorado, New York, Utah, Nevada, Maine and Illinois each have their own AI statutes now. Federal preemption is still unsettled this July, so state law binds you today.
Three specifics for schools. In Texas, TRAIGA's teeth are the ban on using a system built to discriminate unlawfully and the rules on capturing biometric identifiers — face-matching on the yearbook photo library, or a fingerprint reader on the lunch line, needs consent and a policy. The Attorney General enforces, with a cure period, so a documented fix matters.
In Illinois, the employment law that took effect 1 January reaches any school using a tool to screen teaching applicants — including one that ranks resumes for the fourth-grade opening. In New York City, an employer using an automated screening tool needs an annual bias audit and has to notify candidates in advance; a Manhattan independent school hiring a Latin teacher is an employer like any other. Colorado's law names education enrollment as a consequential decision, which puts admissions and aid tools inside it — confirm the effective date with counsel, because it has already moved once.
California SB 53 is aimed at the companies building the largest models, not at your business office — but it is part of why your vendors' contracts changed this spring. Read the amendment your student information system sent you in February. It was not boilerplate.
For a high-risk use, the school's duties are unglamorous and mostly clerical. Use the system the way the vendor's instructions say. Assign a named human with the standing to overrule it — for admissions that is the Director of Enrollment Management, not an associate. Keep the activity records rather than letting them roll off after thirty days. Tell the families, in language a parent reads, that a computer system took part. Watch for it going wrong and write down what you did.
The parent-facing piece is the one schools get wrong. It does not belong buried in the enrollment contract. It belongs in the application instructions and on the financial aid page: what the tool does, that a person makes the decision, and who to ask about it. Two sentences, in your own admissions voice.
Still reading? Stop comparing — try CallSphere live.
CallSphere ships complete AI voice agents per industry — 14 tools for healthcare, 10 agents for real estate, 4 specialists for salons. See how it actually handles a call before you book a demo.
An illustration, not a quote. Assume 340 students, one Registrar, one Director of Enrollment Management, and the counsel you already use for the enrollment contract.
| Item | Assumption | Year-one cost |
|---|---|---|
| Inventory afternoon | Registrar 6 hrs + Enrollment 4 hrs at $48 loaded | $480 |
| Counsel review of the inventory | 2.5 hrs at $340 | $850 |
| Rewriting two parent-facing notices | Communications Director, 3 hrs | $150 |
| Turning on and storing activity records | Tech Director, 4 hrs, no new license | $220 |
| Monthly log review | 1 hr per month, Registrar | $576 |
| Total | $2,276 |
Now the other column. The realistic exposure at your size is not a European fine. It is a denied family whose attorney asks in writing how the decision was made, and a school that cannot produce a record. Reconstructing that decision eighteen months later, from memory and a deleted inbox, costs far more than $2,276 in legal hours alone.
Do not let anything score an application on its own, even as a first pass, even "just to sort." The moment a tool produces a rank order that a tired committee follows at 9 p.m. in early February, you have handed it the decision whatever your policy says. Keep the reading human and let the tool do the clerical work around it: pulling the transcript into the right fields, checking the immunization record in Magnus, flagging a missing recommendation.
Do not put a financial aid file into any general-purpose chat tool. The Parents' Financial Statement, the tax return, the letter explaining a divorce — those are the most sensitive documents in the building after health records.
And do not write your own legal conclusion about whether you are in scope. Write the inventory — it is factual, and nobody else can write it for you — then send it to counsel with one question: which of these rows are high-risk, and why.
You are done with the European part — write the sentence that says so, dated, with the enrollment report attached. You are not done with Texas, Illinois, Colorado or New York City, or with the parent who asks whether a computer read her daughter's essay.
No. The vendor's obligations and the school's are different sets. They build and document the tool; you use it as instructed, name a human reviewer, keep the records and tell the families. A letter saying "we are compliant" does not fill in your columns.
Not high-risk — answering "when is the open house" is not a decision about a person. It does carry a transparency duty: the parent needs to know she is talking to a computer, not the admissions associate. One sentence in the opening message covers it. Keep the transcripts.
Some systems already on the market before the rules applied are treated differently for some obligations. That is a real distinction and a bad plan to build on, because it usually collapses the first time you take a version update. Assume the tools you use this admissions cycle need the file.
One practical note on the phone and the website chat, since both sit in the transparency column rather than the high-risk one. CallSphere builds AI voice and chat agents that answer the school's main line and web chat, book tours and open-house seats, and capture inquiries around the clock — and every call leaves a written record with a timestamp. On the inventory above, an answering agent is an easy row: it announces what it is, it decides nothing about a child, and it hands you the log.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
The EU AI Act's August 2 date, Texas TRAIGA and California SB 53 all landed. What a US rehab clinic must document, disclose and log, and what it can skip.
Private school inquiry calls land in the absence rush and go to voicemail. What 2026 voice agents change, plus the tuition arithmetic and the calls to keep human.
Driver screening tools and cab-facing cameras put carriers under Illinois, Texas and NYC rules in 2026. What to document, and what is genuinely out of scope.
Two casino systems land in the EU AI Act high-risk bucket: face matching and marker scoring. What US gaming operators document, disclose and can ignore.
What a US registered investment adviser must actually document, disclose and log in 2026 — and which AI statutes are genuinely out of scope for the firm.
The EU AI Act's August 2, 2026 date reaches US studios delivering to EU viewers. Here is what to log, disclose and consent for, and what is out of scope.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI