Role guide, version 3.0.1, in force since September 2026

Teach, assess, publish

What your professional ethics already require, what version 3 recommends and supports, and what the law imposes. Three different things, and this guide never blurs them.

This guide completes the common core of the charter. Where it diverges from another internal document, the more protective provision applies. Read the common core.

Scope: Every campus. Permanent faculty, visiting lecturers, doctoral students with teaching duties, affiliated researchers.
Academic freedom: The choice of methods, materials and assessment formats remains yours, within the study regulations and applicable legal obligations.

Build my syllabus block Choose a level

Three levels of commitment

Mistaking a recommendation for an obligation is the surest way to lose the trust of a faculty. So the guide separates them, and says each time where the constraint comes from.

01
Already binding
Professional ethics
Nothing new here. These commitments belong to teaching and scientific ethics. AI only renews the conditions in which they apply.
- Declare your use of AI honestly in your teaching and scientific output.
- Conceal nothing: not in a publication, a course pack, a marking key or a report.
- Check the sources and references AI produces before any circulation.
- Protect student data: no identifying or sensitive information in a tool without an approved framework.
02
Recommended and supported
Teaching standards
The novelty of version 3. The AI Transformation Office supports adoption with training, workshops, templates and shared experience, and adjusts these standards to real use. Nothing is imposed.
- State the expected N0 to N4 level in every syllabus, and repeat it on every assessment brief.
- Build one AI-aware sequence into the course: short, tied to your discipline, where use and limits are worked through explicitly.
- Rethink the assessments most easily substituted: oral defence, tailored case, annotated portfolio, real data.
- Document your own use of AI when preparing classes, for your students and for yourself.
03
Imposed by law
Regulatory obligations
This level does not depend on the charter. It follows from applicable law and scientific integrity. The charter merely spells it out.
- GDPR: no identifying student data sent to an unapproved AI. That includes scripts, learning traces, recordings and private exchanges.
- AI Act, Regulation EU 2024/1689: scoring, ranking or deciding about students triggers the requirements of Articles 12 to 14, auditability, explainability, effective human oversight. Contact the AI Transformation Office before any such deployment.
- Scientific integrity: no fabricated data, no concealed use in publication, respect for authorship rules.

Levels 02 and 03 create no new unnegotiated contractual obligations. Any change to contractual obligations falls to the competent bodies, works council and collective bargaining. Source: faculty guide V3, section 3.

Opting out is allowed, with no consequence

Reasoned opt-out clause

By reasoned decision, you may decline to implement all or part of the level 02 standards: the N0 to N4 statement in the syllabus, the AI-aware sequence, redesigning an assessment, documenting your own use, or attending the training offered by the AI Transformation Office.

The opt-out is exercised within the pedagogical freedom recognised by law, notably Article L. 952-2 of the French Education Code. The AI Transformation Office and the academic registry undertake to apply no pressure, formal or informal.

The level of a task

You set the authorisation level for each assessment. It appears clearly on the brief and is announced at the start of the course. Here is what each level is for, and what it really puts to the test.

The ramp shows how much AI is permitted, from the most closed to the most open. It states an intensity, not an authorisation.
Level	What the student may do	When to choose it	Skill assessed
N0No AI	No AI allowed. Controlled environment.	Calculations, unaided writing, invigilated exams, orals without materials.	Core mastery, automatic skills.
N1Preparation	AI for brainstorming and initial research. Final output written without AI.	Essays, analyses, literature summaries.	Synthesis and personal appropriation.
N2Assistance	AI on specific tasks: structure, rewording, translation. Use statement mandatory.	Reports, case studies, portfolios of work.	Assisted personal output, judgement.
N3Collaboration	Free use of AI. Critical thinking, validation and a statement are required.	Complex projects, dissertations, field assignments.	Critical thinking, tool orchestration, integration.
N4Exploration	Command of AI is itself assessed. Creative and critical use expected.	Innovation projects, hackathons, AI prototypes.	Technological mastery, tooled creativity.

Default values

Work submitted from home

Unless you state otherwise.

N0 or N1

Invigilated exams

Unless stated otherwise.

Final dissertations

With a detailed use statement.

On every brief, state three things: the level chosen, the tools allowed or banned for that task, and the form the use statement should take.

The syllabus block

A frame to paste at the top of a syllabus, or as an annex to a case or an exam paper. Five lines filled in once, and the ambiguity is gone for the semester.

Use of artificial intelligence

ISC Paris AI Charter, levels N0 to N4

Level allowed for this course or assessmentN ___
Tools explicitly allowed
Tools explicitly banned
Use statement expectedyes / no
Why this level, in one sentence

Three formats sit on the AI Transformation Office intranet: Word, Notion and Markdown. The template for the student use statement is in the student guide.

Build the block

Before circulating the syllabus

An N0 to N4 level chosen for every assessment.
One short sentence of justification.
At least one allowed tool listed, except at N0.
The use statement either required, or explicitly not expected.
The block visible on the first page of the syllabus.
The level announced aloud in the first class.

Four blocks already filled in

Written exam on site

No AI tool admitted. Phones and laptops stored away. A non-programmable calculator is tolerated.

Analytical report

Rewording, translation of foreign sources and structuring are allowed. A use statement at the end of the report is mandatory.

Final dissertation

Free use, set out in the methodology: tools, significant prompts, checks performed. An oral defence closes the work.

Hackathon project

AI is the object of the work. The deliverable includes a reflection on the biases and risks of the proposed solution.

Designing robust assessment

The aim is not to eradicate AI, but to design tasks that assess what should be assessed: judgement, understanding, the ability to bring knowledge to bear on a specific context.

Formats that resist substitution

Oral defence
Direct interaction, follow-up questions, immediate verification.
Tailored case study
Unique context, data specific to one company or one field.
Critical reflection
Analysis, argument, a personal position taken.
Timed production
Competence demonstrated in a controlled environment.
Evolving portfolio
The trace of the learning process, not only the finished product.
Work on real data
AI does not know your data. It can help process it, not invent it.

Three briefs that hold

Tailored case
“From the financial data of company X, supplied in the annex, recommend an investment strategy. Defend your choices orally before a panel.”

Critical reflection
“AI produced the attached marketing analysis. Identify its flaws, correct them, and justify each change in a two-page note.”

Portfolio
“Document your learning over eight weeks. Each entry says what you learned, how, what remains unclear, and what you will do to clarify it.”

When calibration becomes the thing assessed

Rather than allowing or banning AI outright, the task tests the student’s ability to know when to use it, when to do without, and why.

Mode-switching competence makes calibration itself the object of assessment. The student works first with AI, then without it, and accounts for what was kept.

2 hours

AI allowed

The student explores, prompts, compares, and notes what was expected from the tool.

2 hours

No AI

They finish alone and justify, at the switch, what was kept and why.

The brief, as it is worded

“For this marketing analysis, you have 4 hours. You may use AI during the first 2, then must finish without AI. At each switch, justify what you expected from the AI and what you kept.”

What you mark

The expectations the student formed about the AI.
The suggestions kept, changed or discarded, and the reasons why.
What is left in the final deliverable that AI alone could not have produced.

“What a grade needs to certify is not the quality of the output but the quality of the judgement behind it.”
Lee, ESCP Summit AI in Higher Education, March 2026

This format suits levels N3 and N4 in particular. It calls for a carefully worded brief and an explicit marking scheme; the AI Transformation Office can help with design and calibration.

Undeclared use that you suspect

Suspicion is settled neither by a detector score nor by conviction. It is settled by a documented body of converging evidence, then by a conversation about the substance.

Multiple signals, never a detector alone

AI detectors produce false positives and false negatives in significant proportions. No detection tool constitutes, as of today, admissible proof on its own.

A detector alone proves nothing

Signals to cross-check, never to read alone

A marked change of style between sections of the same piece.
Vocabulary or phrasing unusually sophisticated for the student’s level.
References outside the syllabus, or non-existent once checked.
Logical inconsistencies hidden behind flawless form.
Excessive detail on minor or irrelevant points.
The absence of the errors typical of learning at that stage.

What the student keeps

They may contest the accusation: adversarial procedure is mandatory.
An AI detector alone never constitutes sufficient proof.
Where disagreement persists, they may refer the matter to the academic registry and the AI Transformation Office.

The charter provides for no sanction of its own. Any consequences follow from the study regulations, applied by the competent bodies.

The procedure, in this order

Analyse
Gather several converging signals, factual and documented.
Talk
Hold a meeting. Questions bear on the substance of the work, the process, the methodological choices.
Document
Keep the factual elements gathered and the student’s explanations.
Decide
Weigh the context, the seriousness, the intent and the study regulations.
Support
A first lapse may call for a teaching response. Sanction is not the first option.

Research

Three families of open uses, five prohibitions that are not up for discussion. The dividing line is simple: AI may help you write and explore, never judge or sign.

Permitted uses

Scientific writing

Language assistance: translation, rewording, style, terminological consistency.
Structuring an article, a chapter, a thesis outline.
Literature summaries, with systematic verification of references.

Analysis

Data exploration, with scripts documented and kept for reproducibility.
Visualisation, with accuracy checked.
Hypothesis generation, subject to scientific justification afterwards.

Research process

Initial state of the art, completed by manual verification of sources.
Peer review on form only: grammar, structure, clarity.

Absolute prohibitions

AI as co-author. Never: it does not sign, it is credited in the methods or the acknowledgements.
Fabricating data, that is presenting fictitious data as real. A major breach of scientific integrity.
Concealing the use of AI in a publication.
Handing AI the peer review of scientific substance: neither the validity of a demonstration nor the relevance of a contribution.
Processing personal data without effective pseudonymisation, prior ethical approval and an approved tool.

True anonymisation, within the meaning of Recital 26 GDPR, is rarely achievable on research data: qualifying it falls to the Data Protection Officer.

Publishing: transparency and reproducibility

Any use of AI in research work is declared in the publication, documented enough to be reproduced, and verified by independent validation.

What the Methods section must contain

The model used and its version.
The main prompts, or the interaction methodology: iterations, refinements.
The limitations identified in the use of AI.
The checks carried out by the authors.

Template statement

“This study used [model, version, provider] for assistance with [specific tasks]. The statistical analyses were carried out by the authors. All bibliographic references were checked manually. The main prompts and intermediate versions are available in the annex, or on request.”

10 years

Recommended retention period, aligned with scientific reproducibility standards.

Keeping the traces

The significant prompts used during the work.
The analysis scripts that involve AI.
Successive versions, where they inform reproducibility.

The AI Transformation Office is working with the research directorate on a shared institutional archive, for faculty who want one.

Before submission, three policies to check

The target journal

Sometimes a dedicated AI policy section.

The publisher

Elsevier, Springer, Wiley and Taylor & Francis hold distinct positions.

The funder

National agencies, European programmes, private funding.

Nature, Science and most AACSB-listed journals have published explicit guidelines, and these requirements change: check at every submission. If a reviewer or an editor asks for your prompts, you must be able to supply them. Prepare the methodological annex while you write.

Research data and ownership

Four regimes not to be confused: what you generate, what you collect about people, what you sign, and what you let a model train on.

Synthetic data

Explicitly labelled as synthetic in every publication.
Transparency on the generation method and parameters.
Assessment of the biases generation may introduce.
Full documentation of the procedure.

Personal data of research subjects

Effective pseudonymisation at the very least, within the meaning of Article 4(5) GDPR.
Prior approval from the competent ethics committee.
Institutionally approved tools, and nothing else.

Your AI-assisted output

You remain the author: standard intellectual property rights apply, subject to the use statement.
Check for plagiarism: AI can reword existing content.
Add references where content derives from identified sources, and document the process.

Training a model on ISC Paris corpora

Written consent from every author concerned: dissertations, theses, articles, course materials.
Compliant hosting, with a preference for the European Union.
Prior institutional approval: AI Transformation Office and research directorate.

Before the first class

The guide comes down to six habits. The seventh is to say it out loud, on day one, in front of the group.

Do	Avoid
Set the AI level for each assessment and write it on the brief	Leaving ambiguity
Read and validate every AI-generated content	Circulating without review
Document your use of AI in research	Concealing AI in a publication
Check sources, references and figures	Trusting citations produced by AI
Choose the right tool for the job	Over-equipping without need
Train students in responsible use	Banning without explaining

The wording to read out in the first class

“This course applies the ISC Paris AI Charter. The authorisation level for this course is N___. That means [...]. A use statement is [mandatory / optional]. If you are unsure what is allowed, I would rather you asked me before submitting your work.”

Annex A2 of the charter, recommended wording for the first class.

A case the charter does not settle

Write to the AI Transformation Office. It supports, it does not police: designing a robust assessment, bringing AI into a course, the stakes of a research project, a use statement before submission, referring a contested case to the competent bodies.

Write to the AITO

Training open to faculty

1 h 35
Teaching with AI
Five micro-modules: capabilities and limits, personal uses, the UNESCO 2024 framework, designing teaching sequences, handling suspected misconduct. Permanent faculty, adjuncts and visiting lecturers.
1 h
Using AI safely at work
The core course for every member of staff: what AI can do, confidentiality reflexes, the N0 to N4 scale, when to declare a use.
45 min
The AI Act: five costly misconceptions
A first step, openly accessible, to place the regulation without entering it.

See the courses on the ISC Paris academy

Article 4 of the AI Act, in force since 2 February 2025, requires any actor deploying AI systems to ensure a sufficient level of AI literacy among the people concerned. ISC Paris meets that obligation through these courses, taken online at your own pace. Each awards a personal certificate after a final quiz passed at 80 per cent: the individual, datable proof that the training has actually been received.