Superintelligence in Schools: What Changes for Leaders and Families

Superintelligence in Schools: What Changes for Leaders and Families

Superintelligence in schools is not something a school buys today. It is a scenario of far more capable AI that forces school leaders and families to decide in advance which data, which actions and which decisions stay with people. In practice, that means governance: clear limits for the front office, for communication with families and for minors' data, with a named owner and human review.

Key takeaways

• Superintelligence is a hypothetical system that outperforms humans at most intellectual tasks. What schools run today are assistants and agents, and the limits for them should be set now.
• The more capable the AI, the more it matters what it can access and do, and the less it matters how intelligent it seems.
• Data on children and teenagers needs extra care under LGPD, Brazil's data protection law. The practical rule is to collect the minimum, restrict access and log every use.
• Communication with families needs a defined channel, permitted topics and a handoff to a person on sensitive subjects such as health, discipline and finances.
• Decisions that affect a student (admission, scholarships, disciplinary measures, grading) should not be made by AI alone.

What does superintelligence mean for a school?

The term describes a hypothetical system that would outperform humans in almost every area of reasoning. It does not exist as a product anyone can buy, and no school needs to make a decision about it. What does exist are language models and agents that are increasingly capable. They already draft notices, answer questions, summarize documents and carry out tasks in connected systems. To understand how the concepts differ, read the article on the difference between AGI, ASI and AI agents.

Why talk about it in a school context at all? Because the public debate about highly capable AI changes the questions that families and school boards ask. Instead of asking whether the tool works, they start asking who is accountable when it makes a mistake, where their child's data is stored and what the school would never let a machine decide. A school that already has written answers is prepared no matter how capable the next model turns out to be.

A useful methodological reference is the NIST AI Risk Management Framework, which organizes the work around governing, mapping, measuring and managing risk. It is not a standard written for schools, but it gives you shared vocabulary for talking with the school's governing body, the board and vendors.

What changes in the school front office?

The front office is where AI proves useful first: copies of documents, enrollment status, the calendar, recurring questions, tracking pending items. With more capable models, the assistant stops merely answering and starts acting: querying the management system, filling in forms, changing statuses, sending messages. That is where risk grows, because a poor answer can be corrected, but a wrong action on a student record can affect a real student.

The central change for the school leader is to separate three permission levels:

• Reading the school's public information (calendar, school rules, list of required documents).
• Reading data on a specific student, only for a verified legal guardian.
• Changing data or starting processes, always with approval from a front office employee.

Each level needs different controls. The first can run with little supervision. The second requires identity verification and a log of every query. The third should never be automatic at the start. Authorization guidance such as OWASP's reinforces the principle of least privilege: the agent only gets the access the task requires.

For an overview of possible workflows, the article on AI for schools: admissions and parent communication explains where to begin.

What happens to communication with families?

In Brazil, much of school communication happens on WhatsApp, in groups, broadcast lists and direct conversations with the front office. More capable AI makes it tempting to automate everything, because it writes well and answers fast. But the right tone is not the same as the right decision, and families tend to treat a message from the school as an official position.

A few principles keep communication under control:

• Define what the assistant may answer on its own: schedules, dates, documents, general guidance.
• Define what it never answers: individual grades outside an authenticated channel, behavior, health, a family's financial situation, disputes between guardians.
• Say clearly that the person is talking to an automated assistant and how to reach a human.
• Log the conversation and the reason for each handoff to staff.
• Send outbound messages only with recorded permission and approved templates, as explained in WhatsApp outbound messages: templates, opt-in and best practices.

There is also a delicate point: the legal guardian is not always the person writing. Grandparents, transport drivers, separated parents and other contacts may use the same number or a different one. Before any individual information goes out, the workflow has to confirm who is on the other end, and the confirmation rule should be set by the school, not by the model.

Minors' data: what should the school limit?

LGPD devotes a specific article to processing data of children and teenagers, Article 14, which requires that it be done in their best interest. For children, the law calls for specific, prominent consent from at least one parent or legal guardian, with exceptions. The exact application depends on the legal basis for each processing activity, so the school should validate each case with legal counsel and its data protection officer. ANPD, Brazil's national data protection authority, publishes guides and guidance that decision makers should treat as required reading.

From a management standpoint, what matters is turning this into operational limits:

• Collect only the minimum needed for the task. If the assistant answers calendar questions, it does not need health data.
• Do not paste student data into open or personal AI tools. The company policy for employee ChatGPT use is a good model for writing this kind of rule.
• Check where the vendor processes and stores data. If processing happens outside the country, the article on international data transfer under LGPD when using AI providers shows what to verify.
• Set a retention period for conversations and records, and decide who can delete them.
• Prohibit using minors' data to train or fine-tune models without a formal decision and a documented legal basis.
• Mask names and identifiers when the goal is only to improve a workflow or test answers.

There is also the matter of automated decisions. LGPD, in Article 20, gives people the right to request review of decisions made solely through automated processing that affect their interests. In a school, that touches admission, scholarship awards, ranking and disciplinary measures. The article on LGPD automated decision-making and the right to review goes deeper. The management rule is simple: AI can prepare, summarize and suggest, but a person decides and is accountable for the outcome.

How do you decide what AI may do? A decision guide

When a new use case appears, the school leader can work through the questions below in order. If the answer to any of them is uncertain, the case goes back for review before any testing.

• Does the task involve personal data of a student or guardian? If so, exactly which fields, and are all of them necessary?
• Is there a recorded legal basis and a named owner for this processing?
• Does the result affect the student's rights or opportunities? If so, AI may only suggest, and a person approves.
• Does the vendor state where data is processed, how long it is kept and whether it is used to train models?
• Is every answer, query and action logged, with date and reason?
• Is there a clear path for the family to reach a person at any time?
• Has anyone tested the workflow with hard questions, such as requests from third parties, attempts to get another student's data and manipulative messages? The OWASP Top 10 for LLM applications lists risks such as prompt injection and sensitive information disclosure that work well as a test script.
• Is there an agreed criterion for stopping or scaling back use if results are poor?

This guide does not replace legal advice or the policy of the school's governing body. It exists so that the conversation does not start with the tool and end with a data problem.

Illustrative scenario: a document request and an off-script question

Illustrative scenario: a mid-sized private school puts an assistant on the front office WhatsApp. It answers questions about the calendar, the document list for re-enrollment and deadlines. When someone asks for an enrollment statement, the assistant confirms the registered contact, opens a request and says the front office will review and send the document.

One day, an unknown number asks whether a specific student was disciplined the previous week and says they are the child's uncle. The assistant looks nothing up: the rule says disciplinary information is never handled in the automated channel. It replies that this topic is handled by a person in the coordination office, logs the request with the reason for the handoff and ends the conversation without confirming or denying anything.

The next day, the coordinator sees the log, checks that the number is not in the guardian records and decides how to respond. The value of this design lies in two points: the limit was written down beforehand, and the log let someone review the case. Neither depends on the model being more or less intelligent.

How does the map, limit, pilot and hand over method apply?

This method, used by Juan Carlo in AI projects for companies, adapts well to schools and does not assume any specific technology.

• Map: describe how the front office and communication work today, which questions repeat, which systems hold student data and who is accountable for each topic. Without this picture, there is no way to know whether AI improved anything.
• Limit: define which data goes in, which actions the assistant can perform and which topics are always sent to people. This is where LGPD, the school rules and the governing body's guidance come in.
• Pilot: choose a single real workflow, such as re-enrollment questions, with logs and a success criterion agreed beforehand, for example less rework in the front office without an increase in complaints.
• Hand over: give the team the documentation, the owners and a review routine, and decide with evidence whether the pilot expands, is adjusted or stops.

The stopping point matters. A pilot the school can end without harm is safer than a system it comes to depend on without understanding it. To keep up with regulatory changes, the article on Brazil AI regulation: what companies should watch and do now helps you keep the agenda current.

Frequently asked questions

Has superintelligence already reached schools? No. What schools use are assistants and agents built on language models, which are capable at specific tasks and fail in unpredictable ways. Treating these systems with access limits, logging and human review is the most realistic preparation for any future development.

Do families need to be told when a school uses AI? Transparency is good practice and fits LGPD's processing principles, such as purpose, necessity and transparency. In practice, it is wise to state in the support channel that the conversation is automated, explain which data is used and show how to reach a person. The exact wording should be validated with the school's legal counsel.

Can a school use ChatGPT or other open tools to handle student data? The prudent recommendation is not to enter identifiable student data into tools that have no contract, no retention controls and no clear definition of how data is used. Staff can use AI for tasks that involve no personal data, such as drafting a general notice, as long as a written policy exists. For minors' data, the vendor should be assessed first, with the support of the data protection officer.

Where should a school leader start? By mapping one low risk front office task, such as questions about documents and the calendar, and by writing the limits before any testing. After that, a short pilot with logs and a clear stopping criterion shows whether it is worth expanding. Topics involving health, discipline and family finances stay out of the initial scope.

Sources

• NIST: AI Risk Management Framework
• ANPD: Autoridade Nacional de Proteção de Dados
• OWASP Top 10 for LLM Applications
• OWASP: Authorization Cheat Sheet