AI Enters the Backpacks of Underage Students
The integration of artificial intelligence in schools has reached a decisive milestone. According to an article published by the New York Times, tech giant Google recently opened access to its conversational agent, Gemini, for underage students using its Google Classroom educational suite. Previously reserved for adult users, this feature is now deployed directly in the daily learning environment of millions of children worldwide. This decision marks a turning point in the penetration of generative AI tools within primary and secondary schools.
This push by Silicon Valley giants responds to a growing demand for writing assistance, personalized tutoring, and exercise generation tools. However, it places school boards and school administrations in front of a major ethical and technical dilemma. How can the adoption of these learning technologies be reconciled with the legal and moral obligation to protect the privacy of minors? Indeed, introducing commercial models directly into classrooms exposes student data to complex, often opaque processing flows located outside national borders.
The Blind Spots of Technological Centralization
The main stumbling block of this massive integration lies in the processing of query data, commonly known as prompts. When a student interacts with a mainstream AI model, the questions asked, the texts submitted for correction, and the academic difficulties expressed are routed to centralized data centres. According to UNESCO guidelines on generative artificial intelligence in education, the risk of commercial profiling of minors and the use of their data to train future models is very real, despite the contractual commitments of providers.
In Quebec, this deployment runs directly into the rigorous requirements of Law 25 on the protection of personal information. According to the Commission d'accès à l'information du Québec, any processing of data concerning minors under the age of fourteen requires the clear consent of the person having parental authority. Furthermore, the law mandates a privacy impact assessment for any transfer of data outside the province. Software solutions from multinationals, designed for a globalized market, rarely offer the granularity and airtight security required to comply with these local legislative specificities.
From an educational standpoint, using unmonitored, general-purpose language models also presents risks of hallucination. Without a precise documentary anchor, these systems can generate incorrect or biased answers, undermining the scientific rigour required in schools. The lack of control over the information sources used by the AI to formulate its answers poses a major problem for academic integrity.
RAG and Sovereignty: A Local Architectural Alternative
In the face of this centralized model, a different approach is emerging in Quebec, demonstrating that it is possible to combine educational innovation with absolute respect for confidentiality. Rather than sending student queries to foreign servers, the ProductivIA platform offers a decentralized, airtight architecture specifically designed for the education sector.
At the heart of this approach is the ÉtudeIA application, a school support tool designed to operate in a closed loop. Unlike mainstream conversational agents, ÉtudeIA does not search for answers across the entire global network. It relies exclusively on the school's or teacher's Document Base. Technically, this process is based on Retrieval-Augmented Generation (RAG). Textbooks, course notes, and approved exercises are converted into vector representations (embeddings). When a student asks a question, the system extracts only the relevant passages from these local documents to formulate a verifiable answer that is free of hallucinations.
To guarantee total sovereignty, the ProductivIA orchestrator allows these queries to be routed to the sovereign Quebec provider Matania. The language models of the Qwen family, which power Matania, are hosted on physical infrastructure located within Quebec. Student data therefore crosses no borders and remains subject to the exclusive jurisdiction of Canadian and Quebec laws. This airtight security is reinforced by the platform's multi-silo architecture, which hermetically isolates the data of each school.
Finally, this software transition is part of a global hardware sobriety initiative. Thanks to Boréal-OS, the ecosystem's native sovereign operating system, schools can give a second life to their aging computer fleets. Computers declared obsolete by Windows 11 requirements can smoothly run this lightweight system and access the ProductivIA platform directly from their browser, without requiring costly hardware repurchases.
Going Further
The arrival of artificial intelligence in classrooms must not come at the expense of students' fundamental rights. The choice facing educational institutions is not a binary one between rejecting technology and technological dependency. By prioritizing sovereign and modular architectures, Quebec schools can adopt these tools of the future while building a solid shield around the personal information of the next generation. The question remains: will public decision-makers choose to prioritize these local solutions to build a resilient educational infrastructure?