Many teachers in France ask the question before trying: are we allowed to grade papers with artificial intelligence? The answer lies in an official document, the framework for the use of AI in education (“L'IA en éducation : cadre d'usage”), published by France's Ministry of Education in June 2025 after consulting the education community. It is not only about students who cheat: it also sets the rules for teachers, grading included. Here is what it says, and what it means in practice.
The principle: AI is allowed, with conditions
The framework states from the outset that “the use of AI is authorised in education provided it respects the framework defined here”. For staff, it specifies that professional use, “administrative as well as educational”, including generative AI services, “is authorised subject to strict compliance with the framework of use”.
Grading is mentioned explicitly. The text notes that some teachers already use AI to “help them with grading and personalised feedback”, and states that for their teaching tasks (“assistance with lesson preparation, assessment, grading, etc.”), any teacher may use AI tools available to the general public, on their own responsibility, provided they check what those tools produce.
The framework's five principles
The ministry sums up the framework in five principles, which apply to grading as to everything else:
- Educational added value: only use AI when it genuinely brings something.
- Data protection, particularly in consumer tools.
- Awareness of environmental impact: the text asks for AI to be set aside if a less environmentally costly solution meets the need.
- Transparency: disclose the use of AI in decision-making.
- Critical thinking: check suggestions, be aware of possible biases and correct them.
Student papers and consumer tools
This is the most important point for grading. The framework is unambiguous about AI services “available to the general public” (free or personal chatbots): their use is authorised “provided that no confidential or personal data is used”. Only data “that can be made public” may be entered into them: official curricula, open resources, anonymised statistics, public-domain works.
A student's paper is personal data: it carries a name, often a class, and its content relates to an identifiable student. The text draws the consequence: consumer services “do not guarantee that the data entered will not be reused”, and they must be avoided “when personal, confidential or copyright-protected data is involved”.
| Content | In a consumer AI tool |
|---|---|
| The exam, a rubric, a model answer | Yes, these are the teacher's documents with no personal data |
| A request for report card comment wording based on an anonymous profile | Yes, if nothing can identify the student |
| A photographed paper with the student's name on it | No |
| A class list, grades attached to names, a report card | No |
The framework also states that no staff member should ask students to create a personal account on these services, and that the use of AI must comply with the GDPR, notably the data minimisation principle and the agreement of the data controller, which, for a school's data, means its leadership.
Final grade, transparency and human oversight
The framework devotes a section to transparency. Using AI in an educational decision must be done “with full transparency and accountability, with explicit communication about its role and the way it was used”. Two rules follow from this, which directly concern assessment:
- any decision made with the help of AI from personal data must be accompanied by information for the people concerned, who must be able to know their rights and act if they disagree;
- any decision relying on AI that has a significant impact on the assessment of learning must be subject to human oversight guaranteeing its fairness and explainability, and to prior approval by the competent authority.
This second rule refers to the European AI Act, which classes systems intended to assess learning outcomes among “high-risk” uses. In practice, a grade generated and copied over without review falls outside the framework; a grade suggested by the AI, checked, adjusted and decided by the teacher, who can explain it to the student, matches what it describes.
What about national exams?
The framework does not deal with the national exams themselves, whose grading follows its own rules. France's Minister of Education, Édouard Geffray, stated in May 2026, in remarks reported by Le Figaro, that grading means “assessing a student's work and the quality of their reasoning with a degree of subtlety”, and that “it is the very essence of the teaching profession”: grading the baccalauréat remains a matter for human examiners.
Classwork, tests, mock baccalauréat exams and mock brevet exams, on the other hand, fall under everyday assessment, where AI assistance is allowed within the framework.
On the student side: cheating and AI detectors
The same text sets the rules for students, which matter to anyone grading. Using generative AI to produce all or part of an assignment “without the teacher's explicit permission and without it being followed by personal work of appropriation” constitutes cheating, treated like the intervention of a third party. Educational use in class is allowed from 4e (the third year of French middle school, around age 13), supervised by the teacher, and independently in high school within a framework the teacher sets.
A point the text flags: software that detects AI-generated content is not recommended, because its lack of reliability “could wrongly penalise a student”. The framework instead encourages teaching teams to adapt assignments, putting reasoning and problem solving first.
In practice: a checklist for grading with AI
- No named papers in a consumer tool. Anonymise them, or use a service that processes data under a contract.
- Check the contractual terms of a professional tool: data processing agreement (DPA), hosting location, no reuse of papers to train a model. Discuss it with your school's leadership and the data protection officer of your académie (France's regional education authority).
- Keep the decision: review every grade, every comment, and correct what needs correcting.
- Be transparent with your students and their families about the help used and the fact that the grade remains yours.
- Only use it when it adds something: a stack of thirty essays, yes; a ten-question multiple-choice quiz, an answer grid is enough.
How Correxio fits this framework
Correxio was designed so that assisted grading stays within these rules, without relying solely on the teacher's vigilance:
- the student's name is hidden automatically on the scan before the AI reads the paper, and you are warned about any paper that could not be masked;
- Correxio acts as a processor within the meaning of the GDPR, with a data processing agreement (DPA) and data hosted in France; papers are not used to train an AI (security and GDPR);
- the AI suggests a score per criterion with its justification, which you review and edit, like every annotation, before exporting anything;
- every grade is based on your rubric, criterion by criterion, which makes it explainable to the student.
In short
France's Ministry of Education allows AI to help with grading, on three conditions: no student personal data in a consumer tool, systematic checking of what the AI produces, and a final decision that remains the teacher's, made transparently. National exams are still graded by humans. Within this framework, AI is a grading assistant, never a grader.