AI has moved quickly into everyday education. According to the EdWeek Research Center, 61% of teachers said they were using AI tools in their work in 2025, compared with 34% in 2023. The shift is happening among students too. A February 2026 Pew Research Center survey found that 54% of U.S. teens had used AI chatbots for help with schoolwork, while 10% said chatbots helped with all or most of their schoolwork.
That puts schools in a difficult position. Banning AI cannot remove it from students’ lives, yet allowing staff and pupils to use random public chatbots can expose school data to unfamiliar–and unsafe–systems. The practical answer sits somewhere in between: choose AI tools that offer useful classroom features while giving the school meaningful control over information.
The starting point should be simple. A privacy-first AI tool should make its data practices understandable, controllable, and appropriate for the age of the students using it.
Start With the Data, Not the AI Features
Most AI product pages lead with what the software can generate. Schools should read the privacy documentation before getting excited about those features.
Find out what happens to every piece of information sent into the system. That includes prompts, uploaded documents, chat history, account information, and technical details collected from the device or browser.
In January 2026, England’s Department for Education revised its safety standards for generative AI products. The guidance states education focused AI products should clearly explain their intended uses and provide information about areas such as data collection, processing, storage, and sharing. It also encourages schools and colleges to consider these standards when assessing AI products.
That gives procurement teams a useful starting point. Instead of asking whether a company calls its product “secure,” ask how long conversations remain stored, who can access them, where processing takes place, and whether information can be passed to another provider.
The answers should apply to the exact version being purchased. A consumer chatbot and an education edition from the same company can have different privacy terms.
Look Beyond the Word “Encryption”
Encryption is valuable, but it only answers part of the privacy question.
A school should also understand what happens after a conversation reaches the provider’s servers. Is chat history retained? How long is deleted information kept in backups? Can administrators delete user accounts along with the content linked to them? And in certain situations, can employees access users’ conversations?
Current Department for Education standards require strong authentication and suitable access controls, with permissions set according to each user’s role. For a classroom system, that could mean teachers receive controls that students never see, while administrators have broader oversight.
It can also be useful to examine how privacy-focused products handle retention. Proton’s current documentation, for example, states Lumo uses a no-logs approach and removes conversation data from its servers after processing, while saved chat history is protected with zero-access encryption. Those are vendor claims, so a school evaluating the product should still examine its contractual terms and suitability for the specific educational setting.
For readers evaluating alternatives outside mainstream platforms, Lumo, a private Gemini alternative, is one example of a service built around privacy as a central product feature rather than an optional account setting.
Check Whether Conversations Feed Model Training
One of the most important questions for schools concerns model training.
Some AI services may use conversations or other submitted material to improve their models, while specialized business or education products can operate under different terms. OpenAI, for example, says data from ChatGPT Edu and ChatGPT for Teachers is excluded from model training by default. Its documentation also notes consumer ChatGPT accounts have separate controls, where users can turn off training for new conversations.
Google makes a similar distinction. Its current guidance mentions people using Gemini through a Google Workspace for Education account receive enterprise-grade privacy and security protections, and chats and uploaded files in Gemini Apps under a school Google account are not reviewed by human reviewers or used to improve generative AI models. Google also stresses that data handling depends on the Workspace license. Personal Google Accounts follow different rules.
This is where many schools can make costly mistakes. Seeing a familiar AI brand on a staff member’s personal account tells you very little about whether the same service is suitable for student records.
Minimize Student Data Before It Reaches the Model
Even a strong privacy product should receive only the information it needs.
A teacher asking an AI system to improve feedback rarely needs to include a student’s full name, address, phone number, or detailed personal history. Replacing identifying information with a neutral label can achieve the same instructional goal while reducing exposure.
The U.S. regulatory framework has its own protections for children. The Federal Trade Commission notes schools can sometimes consent on behalf of parents when an online service collects children’s information for the benefit of the school and within an educational context. That authority comes with limits where information is collected or used for an operator’s separate commercial purposes.
Privacy Should Never Come at the Cost of the Lesson
A beautifully written privacy policy cannot rescue a weak classroom product.
Teachers need AI tools that can support actual learning activities. That could mean helping a student understand a difficult passage, giving feedback on a draft, creating differentiated material, or helping a teacher adapt an existing lesson.
The important question is how the AI handles the student’s work.
A useful educational system can encourage the learner to attempt a problem before revealing a solution. It can provide hints when a student gets stuck. It can ask the learner to reconsider an answer instead of immediately producing a replacement.
Test Administrative Controls Before Buying
A teacher may have little control over what happens to student data if the platform lacks a proper administration layer.
Look for controls that allow authorized staff to manage accounts and determine which features students can access. Age restrictions can also be important. Google’s current Gemini education documentation, for example, notes that some features have age based access requirements.
Run a Small Classroom Trial First
A short pilot can reveal problems that a vendor demonstration hides.
Use fictional student information during the trial. Have teachers perform ordinary classroom tasks and then test a few situations involving sensitive information. Watch what data the system requests, how conversation history behaves, what administrative controls are available, and whether students can access features they should not.