AI in Brazilian Schools: New Rules Set Limits on the Use of Artificial Intelligence by Students and Teachers

The discussion around artificial intelligence in Brazilian schools is entering a new phase.

During the first years of generative AI adoption, much of the educational debate revolved around a relatively simple question:

Can a student use artificial intelligence to complete an assignment?

That question remains important.

But it is no longer enough.

As of September 2, 2026, the Brazilian debate increasingly involves broader questions:

  • should children use AI without supervision?
  • can an AI system grade an exam?
  • can schools send student data to generative AI models?
  • can automated systems evaluate behavior?
  • should teachers disclose when they use AI?
  • how can schools distinguish work produced by a student from work produced by a machine?
  • what role should AI play in the learning process?

Brazil's National Council of Education (CNE) has advanced discussions around national guidelines for the use of artificial intelligence in education, based on principles such as human oversight, protection of children and adolescents, transparency, and risk classification.

One of the most relevant issues concerns early childhood education and the first years of elementary school.

The regulatory logic is increasingly clear:

The younger the student and the greater the impact of the decision, the stronger human oversight should be.

This may appear to be an issue limited to schools.

It is not.

It anticipates a debate that companies will also face in the coming years.

The fundamental question will be the same:

Which decisions can we delegate to artificial intelligence, and which ones must remain under human responsibility?

Brazil is moving beyond the "can students use ChatGPT or not?" phase

When generative AI tools became popular, schools faced an immediate problem.

A teacher assigned an essay.

The student entered a few instructions into an artificial intelligence system.

Seconds later, a complete text appeared.

This produced understandable reactions.

Some institutions tried to ban the technology.

Others ignored it.

Others began experimenting.

But the discussion evolved.

Today, artificial intelligence can participate in virtually every stage of the educational process.

Lesson planning
        ↓
Material creation
        ↓
Personalized explanation
        ↓
Exercises
        ↓
Grading
        ↓
Assessment
        ↓
Performance analysis
        ↓
Pedagogical recommendation

This means we are no longer talking only about a tool used by students.

We are talking about a possible AI-based educational infrastructure.

The first major distinction: using AI and delegating decisions to AI are not the same thing

This distinction is fundamental.

Imagine two scenarios.

Scenario A

A teacher uses artificial intelligence to generate ten ideas for exercises about the French Revolution.

The teacher reviews the suggestions.

Selects five.

Corrects errors.

Adapts them for the class.

Publishes the activity.

In this case:

AI
↓
Suggestion
↓
Teacher
↓
Analysis
↓
Decision

Now consider another scenario.

Scenario B

An artificial intelligence system automatically analyzes a student's answers and determines that the student has poor academic performance.

Based on that conclusion, the system lowers the student's grade or produces an academic recommendation.

Now we have:

Student
↓
Data
↓
AI
↓
Assessment
↓
Consequence

Both cases involve artificial intelligence.

But the level of risk is completely different.

Risk-based thinking is entering education

This is an important shift.

Instead of simply asking:

"Is AI involved in this process?"

the question becomes:

"What is the impact of AI in this process?"

A tool used to generate activity suggestions has one level of risk.

A system used to determine a child's grade has another.

A possible risk matrix could look like this:

AI Use Potential Risk Oversight
Generate exercise ideas Low Teacher review
Summarize content Low Review recommended
Personalize activities Medium Teacher supervision
Grade objective questions Medium Audit
Grade subjective answers High Human review
Assess behavior High Strong controls
Apply disciplinary sanctions Very high Human decision
Analyze biometric data Very high Strict controls

This logic brings education closer to broader discussions around artificial intelligence governance.

Children require an additional layer of protection

There is an important difference between implementing artificial intelligence in a company and making it available to a child.

Adults may not fully understand how an AI model works either.

But children are still developing.

This includes:

  • cognitive development;
  • critical thinking;
  • autonomy;
  • writing skills;
  • argumentative ability;
  • source interpretation;
  • social development.

For that reason, simply giving a child access to artificial intelligence and saying:

"Use it to learn."

may be pedagogically insufficient.

The risk of cognitive outsourcing

There is an important concept that helps explain this problem.

We can call it cognitive outsourcing.

Consider an assignment.

The teacher asks:

Explain why the French Revolution occurred.

The educational objective is not merely to obtain a correct answer.

The student needs to:

  1. read;
  2. interpret;
  3. organize information;
  4. establish relationships;
  5. formulate an explanation;
  6. write.

Now imagine this process:

Question
↓
Copy
↓
Paste into AI
↓
Answer
↓
Copy
↓
Submit

The final result may be correct.

But the cognitive process disappears.

That is the central issue.

In education, the final product is not the only objective. The process of reaching it is also part of learning.

A calculator offers an interesting comparison

This discussion is not entirely new.

When calculators entered classrooms, similar concerns emerged.

If a calculator can solve:

728 × 47

why should the student learn multiplication?

The answer lies in the educational objective.

At a certain stage, the student needs to understand the process.

Later, the tool can increase productivity.

AI can be viewed in a similar way.

BEFORE DEVELOPING THE SKILL

AI replaces effort
↓
May harm learning


AFTER DEVELOPING THE SKILL

AI expands capability
↓
May increase productivity

The difficulty lies in identifying when that transition occurs.

Should students learn how to write before delegating writing?

This question will become central in the coming years.

Imagine a child learning to write essays.

The student needs to develop:

  • vocabulary;
  • structure;
  • argumentation;
  • cohesion;
  • creativity;
  • revision skills.

If artificial intelligence produces the text from the beginning, part of that development may be weakened.

But for a more advanced student, the same AI can function as:

  • an editor;
  • a discussion partner;
  • a tutor;
  • a generator of counterarguments;
  • a research tool;
  • a structuring assistant.

Therefore, the problem may not be:

whether to use AI or not.

But rather:

at what stage of development it should be used and for what purpose.

The teacher cannot disappear from the process

One of the central ideas in current regulatory discussions is preserving human responsibility.

This is particularly important in education.

AI can identify patterns.

But teachers may know things that are not present in the data.

Imagine a student's performance dropping for two weeks.

The system sees:

Lower grades
+
fewer activities completed
+
more mistakes
=
declining performance

The teacher may know:

The student's parents are going through a divorce, and the student is experiencing emotional difficulties.

That information completely changes the interpretation.

Artificial intelligence works with available data.

Teachers also work with human context.

AI should not turn education into a metrics factory

There is another risk.

The more digital systems enter education, the more data can be collected.

For example:

Reading time
Number of mistakes
Response speed
Login time
Number of attempts
Participation
Grades
Attendance
Behavior

With artificial intelligence, all of this data can be correlated.

That may create extremely useful tools.

But it can also create an excessively quantitative view of the student.

A child is not simply:

Student #29481
Performance: 71%
Engagement: 63%
Academic risk: 42%

Education contains dimensions that are difficult to convert into numbers.

Educational data is extremely sensitive

This issue is directly connected to digital security.

Imagine an educational platform containing:

  • name;
  • age;
  • address;
  • grades;
  • learning difficulties;
  • academic history;
  • behavior;
  • family information;
  • potentially biometric data.

Now imagine that this data is sent to different artificial intelligence systems.

The architecture can become:

SCHOOL
  ↓
Educational platform
  ↓
API
  ↓
AI model
  ↓
Another vendor
  ↓
Cloud

Each new connection expands the governance surface.

LGPD still applies when artificial intelligence is involved

There is a common mistake in AI projects.

Companies and institutions sometimes treat artificial intelligence as though it existed outside the legal frameworks already in place.

It does not.

If a system processes personal data, Brazil's General Data Protection Law — LGPD remains relevant.

In education, the issue becomes even more sensitive because the data often concerns children and adolescents.

That means schools need to ask:

  • what data is being collected?
  • why is it being collected?
  • where is it stored?
  • who has access?
  • which vendors receive the data?
  • how long will it be stored?
  • is the data actually necessary?

The problem of copying and pasting data into AI

Imagine a teacher using a public artificial intelligence tool.

The teacher wants help analyzing class performance.

So they paste:

João da Silva — grade 4.2
Maria Santos — grade 8.7
Pedro Souza — grade 5.1
Ana Oliveira — grade 9.3

and send it to the system.

It may seem harmless.

But there is a fundamental question:

Was this tool authorized to receive that personal data?

This is where governance becomes part of everyday operations.

The security problem is not always a sophisticated hacker.

Sometimes it is simply an employee using a useful tool in the wrong way.

Shadow AI also exists in schools

Companies already face a phenomenon known as Shadow AI.

It occurs when employees use artificial intelligence tools without formal approval or organizational awareness.

For example:

Company approves AI A

Employee uses AI B
Employee uses AI C
Employee installs extension D
Employee connects agent E

Something similar can happen in schools.

Teachers and staff may use:

  • chatbots;
  • presentation generators;
  • automated grading tools;
  • image generators;
  • transcription tools;
  • browser extensions;
  • educational platforms.

Each tool may receive data.

That is why simply saying:

"Our school uses artificial intelligence."

is not enough.

The institution needs to know which artificial intelligence systems are actually being used.

Schools will need AI policies

In the coming years, institutional AI policies may become as common as policies related to:

  • security;
  • privacy;
  • internet use;
  • devices;
  • assessment.

A basic policy could answer:

Which tools are allowed?
        ↓
What data may be submitted?
        ↓
Which activities may use AI?
        ↓
When must students disclose AI use?
        ↓
When must teachers review the output?
        ↓
Which decisions cannot be automated?
        ↓
How will incidents be reported?

This transforms AI from an informal experiment into a governed process.

A complete ban will probably not be enough

There is an understandable temptation:

"Let's simply ban artificial intelligence."

But this creates practical problems.

First, because AI is increasingly built into tools students already use.

Second, because students will need to learn how to work with AI in the labor market.

Third, because prohibition does not necessarily mean the technology will not be used.

It may simply mean that its use becomes hidden.

The scenario becomes:

BAN
↓
Student continues using AI
↓
Hidden use
↓
Teacher cannot provide guidance
↓
Less education about risks

For that reason, digital education may be more effective than absolute prohibition.

The goal should be AI literacy

For decades, literacy mainly meant learning how to:

  • read;
  • write;
  • interpret.

Then we added digital literacy.

Now a new competency is emerging:

artificial intelligence literacy.

This means understanding at least basic concepts such as:

  • models can make mistakes;
  • answers can be fabricated;
  • AI does not "know" things in the human sense;
  • prompts influence outputs;
  • submitted data may have privacy implications;
  • results need verification;
  • generated content must be used ethically.

Knowing how to ask is not enough

There is a great deal of discussion around prompt engineering.

But AI literacy cannot be reduced to:

"Learning how to write good prompts."

A truly capable user needs to know how to:

ASK
+
EVALUATE
+
VERIFY
+
QUESTION
+
EDIT
+
DECIDE

The most important skill may not be obtaining a good answer from AI.

It may be recognizing when that answer is wrong.

Hallucinations take on a different meaning in the classroom

Generative models can produce incorrect information in a highly convincing way.

This phenomenon is often called hallucination.

Imagine a student asking:

Who signed a particular historical treaty?

The AI responds with:

  • a name;
  • a date;
  • context;
  • an explanation.

Everything looks perfect.

But part of the information was invented.

A student without prior knowledge may not notice.

This creates an educational paradox:

To use a tool that provides answers effectively, students need enough knowledge to challenge the tool's own answers.

Sources become essential again

For a long time, search engines trained users to expect a list of sources.

Chatbots changed that experience.

Instead of:

Question
↓
10 links
↓
User researches

we now have:

Question
↓
Ready-made answer

That is extremely convenient.

But it can remove an important step: investigating where the information came from.

In education, teachers will need to reinforce questions such as:

  • what is the source?
  • who published it?
  • when?
  • is there evidence?
  • do other sources confirm it?
  • does the cited reference actually exist?

Assessments will also need to change

If artificial intelligence can produce a traditional essay in seconds, assessments based only on content delivery may lose some of their ability to measure learning.

This does not mean abandoning written assignments.

It means redesigning them.

For example:

Traditional model

Write 2 pages about the Industrial Revolution.

Model adapted for the AI era

1. Produce an initial explanation.

2. Ask an AI system to provide three different interpretations.

3. Identify errors or oversimplifications.

4. Compare them with two academic sources.

5. Explain which arguments you accepted or rejected.

6. Defend your conclusion orally.

Now artificial intelligence becomes part of the activity.

But it does not replace thinking.

The process may matter more than the final answer

This shift can transform education profoundly.

Instead of evaluating only:

"What was the answer?"

the teacher can evaluate:

"How did you reach that answer?"

This encourages:

  • reasoning;
  • research;
  • argumentation;
  • verification;
  • authorship;
  • critical thinking.

Paradoxically, artificial intelligence may force schools to value human abilities even more.

Teachers will need training before students

It is unreasonable to expect teachers to guide students on a technology they themselves have not had the opportunity to understand.

Teacher training will be essential.

A professional development program could include:

MODULE 1
How generative models work

MODULE 2
Prompting and limitations

MODULE 3
Hallucinations and verification

MODULE 4
Privacy and LGPD

MODULE 5
Assessment in the AI era

MODULE 6
Ethics and authorship

MODULE 7
Educational tools

MODULE 8
Digital security

The goal is not to turn every teacher into an artificial intelligence engineer.

It is to provide enough competence to use and supervise the technology responsibly.

AI can be extremely useful when applied properly

A risk-based approach does not mean assuming artificial intelligence is harmful to education.

There are highly valuable applications.

A teacher can use AI to:

  • adapt an explanation;
  • create exercises;
  • generate examples;
  • prepare activities;
  • compare approaches;
  • simplify texts;
  • create accessible material;
  • produce questions;
  • simulate debates.

A student can use it to:

  • ask for a different explanation;
  • practice languages;
  • review concepts;
  • receive practice questions;
  • simulate an interview;
  • test knowledge.

The point is simple:

AI works better as an amplifier of the educational process than as a substitute for the educational process.

Imagine AI as a Socratic tutor

Instead of immediately providing the answer, an educational system could ask:

What do you already know about this problem?

Then:

What would your first hypothesis be?

Then:

What evidence could confirm that hypothesis?

Then:

Is there an alternative interpretation?

In this model:

Student thinks
↓
AI asks
↓
Student responds
↓
AI challenges
↓
Student reconsiders
↓
Learning

Artificial intelligence does not replace reasoning.

It stimulates it.

This also matters to technology companies

Educational regulation creates opportunities for technology companies.

Schools will need systems that provide:

  • access control;
  • identity management;
  • logs;
  • auditing;
  • data protection;
  • oversight;
  • secure environments;
  • tool control;
  • usage policies.

It is no longer enough to say:

"We have AI."

The market will increasingly demand:

"We have governable AI."

EdTech companies will need to think about compliance from the product level

A company developing educational technology needs to consider from the beginning:

Product
  ↓
Data collected
  ↓
Purpose
  ↓
Consent / legal basis
  ↓
Security
  ↓
AI model
  ↓
Decision
  ↓
Audit

Adding governance after the product is already built is usually more difficult.

That is why concepts such as:

Privacy by Design

and

AI Governance by Design

are becoming increasingly important.

In other words:

privacy and governance need to be part of the architecture from the beginning.

The school debate anticipates the future of business

There is an interesting connection.

Today we ask:

Can AI determine a student's grade?

Tomorrow companies will ask:

Can AI decide who gets hired?

Or:

Who receives credit?

Or:

Which customer gets blocked?

Or:

Which employee gets promoted?

The ethical structure is similar.

AI produces recommendation
        ↓
Is there a significant impact?
        ↓
YES
        ↓
Human oversight
        ↓
Responsibility
        ↓
Audit

Education is becoming one of the first major social laboratories for this discussion.

Human oversight cannot be merely symbolic

There is another important detail.

Putting a human into the process does not automatically solve the problem.

Imagine:

AI analyzes 10,000 cases
↓
Human receives recommendation
↓
Clicks "approve"

Technically, there is human oversight.

In practice, there may be no meaningful analysis.

This is sometimes described as automation bias: the tendency to overtrust recommendations from automated systems.

Therefore, oversight needs to be meaningful.

The human needs:

  • information;
  • authority;
  • time;
  • the ability to challenge the system.

We will need to learn how to disagree with machines

This may be one of the most important skills of the artificial intelligence era.

For decades, teaching technology meant learning how to use it.

Now we will also need to teach:

when not to obey it.

A student needs to be able to say:

This answer is wrong.

A teacher:

This recommendation does not consider the context.

An employee:

This automation should not execute this action.

A manager:

This model does not have enough evidence to make this decision.

That is digital maturity.

The future will not be "human versus AI"

Much of the public debate still relies on a simple opposition:

HUMAN
VS
ARTIFICIAL INTELLIGENCE

But the more likely scenario is:

HUMAN
+
ARTIFICIAL INTELLIGENCE
+
RULES
+
PROCESSES
+
RESPONSIBILITY

The question will not be whether we use AI.

It will be how authority is distributed between humans and automated systems.

What can schools do now?

Even before all regulations are finalized, institutions can begin with practical measures.

1. Map the tools

Identify which AI systems are being used by teachers, staff, and students.

2. Classify data

Define which information must never be entered into public AI tools.

3. Create age-based rules

An eight-year-old should not necessarily have the same level of technological autonomy as a seventeen-year-old.

4. Review assessments

Identify activities that can be completely outsourced to AI.

5. Train teachers

Not only in tools, but also in:

  • ethics;
  • security;
  • privacy;
  • verification.

6. Require transparency

When appropriate, students can disclose:

Tool used:
Purpose:
Main prompt:
How the answer was verified:
What was changed by the student:

7. Preserve human decisions

High-impact decisions require effective human oversight.

Schools may need to teach AI the same way they teach source evaluation

There is an important educational opportunity here.

Instead of treating artificial intelligence only as a threat to authorship, teachers can turn it into an object of analysis.

For example:

Ask the AI about a historical event.

Then:

Identify three claims.

Then:

Verify each one using reliable sources.

Then:

Explain where the answer was accurate and where it oversimplified the topic.

Now AI becomes a laboratory for critical thinking.

The goal is not to create AI-dependent users

Poor technology education may produce people who know how to ask a machine everything.

Good technology education should produce people capable of deciding:

when to use it;

how to use it;

how to verify it;

when to distrust it;

and when simply not to use it.

That difference will become increasingly important.

Artificial intelligence is forcing education back to its fundamental question

There is an interesting irony.

The more capable machines become at producing answers, the more important it becomes to ask:

What does it mean to learn?

If learning simply meant producing a correct answer, AI would solve much of the problem.

But education involves:

  • understanding;
  • memory;
  • reasoning;
  • judgment;
  • creativity;
  • interpretation;
  • dialogue;
  • human development.

That is why simply replacing tasks with automation does not necessarily represent educational progress.

Conclusion

The regulation of artificial intelligence in Brazilian schools represents much more than a debate about educational technology.

It raises a fundamental question for society:

How much authority are we willing to give artificial intelligence systems?

In education, this question becomes particularly sensitive because we are dealing with people who are still developing.

Artificial intelligence can help teachers.

It can personalize explanations.

It can expand access.

It can reduce administrative work.

It can create new learning experiences.

But it can also:

  • encourage dependency;
  • expose data;
  • reproduce errors;
  • automate judgments;
  • weaken cognitive processes;
  • create decisions that are difficult to audit.

The most mature response will probably be neither complete prohibition nor unrestricted use.

It will be building governance.

That means combining:

INNOVATION
+
OVERSIGHT
+
SECURITY
+
PRIVACY
+
TRANSPARENCY
+
ACCOUNTABILITY

And there is a lesson that extends beyond school walls.

The children learning how to use artificial intelligence today will become the professionals who tomorrow work alongside autonomous agents, automation systems, and generative technologies.

Teaching only how to use AI will not be enough.

We will need to teach something more difficult:

how to keep thinking when there is always a machine ready to think for you.

That may be one of the greatest educational challenges of the artificial intelligence era.

References

  1. Brazilian National Council of Education (CNE) / Ministry of Education. Acts, opinions, and regulatory discussions concerning artificial intelligence, digital education, and the use of technology in basic education.

  2. Brazilian Ministry of Education (MEC). Guidelines and measures related to digital education, curriculum integration, and Brazilian education policy.

  3. Bahia State Department of Education. SEC establishes guidelines for the use of Artificial Intelligence in the classroom. Published February 21, 2026.

  4. Secretariat for Social Communication of the Presidency of Brazil. Initiatives related to Digital Education and AI curricula and media literacy.

  5. Law No. 13,709/2018. Brazilian General Data Protection Law — LGPD.

  6. Law No. 14,533/2023. Brazilian National Digital Education Policy — PNED.

  7. CNE/CEB Resolution No. 2/2025. National Operational Guidelines on the use of digital devices in school environments and the curricular integration of digital and media education.

Editorial notice: this content is informational and analytical in nature and does not replace specific legal, regulatory, or pedagogical advice. Rules approved by collegiate bodies may depend on ratification, publication, or subsequent regulation before producing all of their legal effects.

Content produced for the Vórtice Blog — technology, artificial intelligence, security, regulation, automation, and digital transformation.