Discovery training
A first session to give the whole team the same foundation: what these tools can do, what they cannot, and the first few moves that save time from the next morning on.
Training by role
AI training for in-house counsel and lawyers in Switzerland
Train your lawyers and in-house counsel in an AI practice as rigorous as it is fast, without ever compromising on the confidentiality of case files
- More than 150 organisations supported across French-speaking Switzerland and France
- 4,500 professionals trained since January 2023
- 5.0/5 on Google (51 reviews)
- Maximum of 12 participants per session, at your offices or remotely
- Training delivered in French and English, German available
12participants max per session
An in-house counsel or a lawyer who tries ChatGPT runs into two problems quickly. The first: the tool invents references with the same confidence it shows when it is right. The second, more serious in your profession than in any other: knowing where the case data goes the moment it is pasted into a prompt. Our AI legal training starts there: embedding AI into the day-to-day processes of legal teams, with a written rule on what may leave the file.
As a result, many firms and legal departments move forward in scattered order. Some colleagues already use artificial intelligence daily, on their personal account, with case data that should never go there. Others forbid themselves any use out of caution, and lose considerable time on tasks AI dispatches in minutes.
Our position: AI has its place in legal practice, provided confidentiality is treated as a non-negotiable prerequisite and hallucination as a professional risk to be managed, not as a technical detail. The question is never "is AI reliable", but "what does an undetected error cost on this file".
The cost of waiting
AI legal training: what it really costs
Without a clear framework, a legal function loses on both counts: no time saved anywhere, and a data leakage risk nobody is steering.
Legal and case-law monitoring
Tracking how a text evolves, finding comparable case law, checking that a standard clause still holds: this groundwork is still done by hand, file after file, when a properly framed deep research run does it in minutes.
Repetitive drafting
Contracts, formal notices, opinions: much of the structure repeats from one file to the next. Starting from a blank page every time means losing time on form at the expense of substance.
Summarising bulky files
Reading a full set of articles of association, months of email exchanges or a bundle of exhibits before a hearing takes hours. AI can prepare a first summary, provided it is framed so that it does not invent what it cannot find.
Poorly controlled confidentiality
The real risk is not AI itself, it is unsupervised use. A Qualinsight study of 1,030 working people confirms it at national scale: 72% of professionals already use AI at work, 79% have never been trained, and 70% have never checked where their data goes. An employee pasting an extract from a client file into a consumer version of a chatbot is a potential data leak nobody in the firm approved or even saw happen.
Undetected hallucinations
A language model states a false legal reference with the same confidence as a true one. Without a systematic verification protocol, the error can end up in a deed, an opinion or a pleading before anyone spots it.
Meanwhile, the organisations that have set a clear framework already save several hours per week per person, without sacrificing an ounce of rigour.
Since January 2023
AI legal training: numbers, not promises
- More than 150 organisations trained across French-speaking Switzerland and France
- 4,500 professionals trained since January 2023
- 5.0/5 from 51 Google reviews
- Sessions delivered in French, English and German
5.0 out of 51 Google reviews
Train your teamsAI use cases in legal practice
Four families of tasks, and for each of them what AI genuinely clears in minutes. None of these tasks finishes without a lawyer, all of them start faster. On each one, AI delivers a first version and the lawyer verifies it: AI states a fabricated case-law reference with exactly the same confidence as a real one, so human checking of every reference, every citation and every date is not negotiable. Maijin is a training organisation, not a law firm: we issue no legal opinions and professional responsibility for your output remains entirely yours.
| Area | Use cases worked through in the session |
|---|---|
| Drafting |
|
| Research |
|
| Analysis |
|
| Compliance |
|
Our method
AI training for Legal in Switzerland: we come to where your teams are
Training built on your real files, not on generic examples
Before each session we ask you for three or four genuine pain points: a file that keeps slipping, a document nobody has time to write, a manual step everyone copies from the last person. The exercises are built on those, never on a textbook case.
From the first hour we also set a clear rule about what may go through an AI tool and what must never leave your organisation. Your teams walk away with the habit, not just with the demo.
Our approach
We train legal professionals to put generative AI to work
We are not the vendor of yet another legal tool. We train your teams to use the tools you already have, or are about to choose, with a method that puts confidentiality and source verification ahead of speed.
Our approach:
Step {n} 01
1. Confidentiality and sovereignty first, on three levers
The contractual lever first, the one that settles most of the risk for a modest monthly cost: a professional version with a data processing agreement, zero data retention and no training on your data (two tools approved in most of the charters we write: Microsoft Copilot and Claude, never a free consumer version). The technical lever next: data hosted in Switzerland or the EU where available, single sign-on, fine-grained access rights. The classic trap: rolling out a paid tool without first checking who has access to what in your document systems. The human lever last: classifying case data into four levels, and a short written usage charter so that every colleague knows where the line is. On the Cloud Act: the risk is real, but largely mitigated by combining a data processing agreement with zero data retention. If sovereignty is imperative for part of your caseload, the answer is not all or nothing but routing between a local or Swiss tool for routine work and the cloud for complex work.
Step {n} 02
2. The three anti-hallucination rules, applied systematically
Before any document summary, we train your teams to frame the AI with three simple instructions: flag missing or uncertain information explicitly rather than inventing it, quote for every statement the exact extract from the source document that supports it, and never draw on knowledge outside the document provided. AI stays a fast intern, never the source of truth.
Step {n} 03
3. Support in spaced half-day sessions
Half-day sessions spaced 1 to 3 weeks apart, a rhythm compatible with the deadlines of a legal department. Between sessions, exercises on your real files (anonymised if needed), we correct the prompts and adjust the content to your actual cases.
This is AI legal training built for teams bound by professional secrecy: your lawyers and counsel save time on drafting and research, without ever losing control of what goes into a prompt or of the verification of what comes out.
We have worked with organisations with high confidentiality requirements, in private banking, pensions, insurance and the public sector. Two engagements with an explicit legal or governance component: for a Geneva foundation in the financial sector, legal and compliance monitoring together with the structured extraction of data from public reports; for a network of community health and social care centres in a French-speaking Swiss canton, the governance of AI use and compliance with Swiss data protection law for a practice that had been informal until then.
Our AI legal training formats
Three ways in, depending on your team's starting level and the time you can free up for them.
Spaced half-day support programme
The complete path, from the foundation through to the most advanced uses in your field, delivered as half-day sessions spaced 1 to 3 weeks apart. A rhythm designed for busy managers: in between, your teams work through exercises on their real files, and we adjust the next session around whatever got stuck.
Highly specific workshop
One short slot, one tool, one precise task. The format to pick when the foundation is already there and a single use case still needs unlocking.
Example
Claude to spot risky clauses in a supplier contract, as in-house counsel
90-minute workshop: CHF 1,000. Half day: CHF 2,000. Full day: CHF 3,000 to 3,500. Day of targeted workshops: CHF 2,500. Rates exclude VAT, apply per group session of up to 12 participants, and include travel costs. The spaced half-day programme is quoted on the half-day rate (CHF 2,000 per session).
Our track record in artificial intelligence training in Switzerland
+150
organisations accompagnées
4 500
professionnels formés
79 %
des collaborateurs déclaraient perdre du temps sur des tâches répétitives, avant l'audit IA mené pour un réseau de centres médico-sociaux d'un canton romand
4
centres médico-sociaux valaisans réunis autour d'une charte IA commune à l'issue de notre mission
Scoping call, no commitment
Let us find the AI legal training format that fits your team
30 minutes to scope your need, quote within 48 hours.
Discuss your projectNo commitment, and the scoping call is not charged.
Études de cas : l'IA juridique et la conformité dans des organisations suisses
Cas 1 : un réseau de centres médico-sociaux romand, de l'usage caché à la gouvernance de l'IA
Contexte
Un audit qualitatif et quantitatif a montré que 79 % des collaborateurs déclaraient perdre du temps sur des tâches répétitives et que 75 % utilisaient déjà l'IA hors de tout cadre. La phase de gouvernance co-construite qui a suivi a produit une charte collaborateurs ramenée de 7 à 3 pages, une fiche réflexe d'une page validée à l'unanimité et une classification des données en 4 niveaux.
Cas 2 : une fondation genevoise du secteur financier, veille juridique et extraction de données
Contexte
Une veille juridique et compliance outillée avec des templates de prompts réutilisables, et l'extraction structurée de 61 champs depuis chaque rapport public, restitués en JSON directement exploitable puis versés dans la base de données de la fondation.
Ils nous ont fait confiance
CPEG
Rentes Genevoises
OCAS Genève
MBC
RTS
Manotel
FOVAHM
Vicario Consulting
CVAJ
Shiseido EMEA
AI legal training near you: Geneva, Vaud, Valais, Neuchâtel, Fribourg
We travel to your premises across French-speaking Switzerland. Each canton page details the formats, the prices and the companies we have trained there.
Your AI trainers for Legal
Jean-Baptiste Berthoux
Chief AI Officer
An engineer by training, Jean-Baptiste gave his first AI coaching in January 2023. Drawing on his experience, he developed a unique pedagogy, the fruit of his field work and his mastery of the tools.
LinkedIn
Timo Schuler
Consultant and Trainer
For over ten years, Timo has worked at the same intersection: human organizations and digital tools. At Nestlé's international headquarters, he contributed to global HR analytics and the adoption of digital solutions. At Swiss Post, as People Development Innovation Scout, he conducted parallel cognitive science research at the University of Fribourg on the use of AI in employee development. A daily AI practitioner since 2023, he now supports teams and executives in consulting, training and implementation, in Switzerland and France. Languages: French · English · German · Swiss-German
FAQ: AI training for Legal in Switzerland
Yes, and that is the most common case. No technical prerequisite is required: a computer and an internet connection are enough. Groups are capped at 12 participants and everyone works on their own files, anonymised where needed, which lets us set the pace by the team's actual level rather than by a theoretical syllabus. Two weeks before the session, each participant receives a scoping questionnaire and a setup sheet for the tools. That is what lets us hold to the rule of zero installation on the day: training time is spent working on your files, not unblocking accounts. For firms and legal departments working on case documents, one further condition applies: we train exclusively on professional versions, never on free consumer versions, and we check beforehand that this is compatible with your IT charter and your data protection obligations. At the end of the training you leave with the materials in PDF, a prompt library tailored to your tasks and summary sheets.
During the training we access none of your case files: your teams work on their own documents, anonymised where needed, and we keep nothing after the session. On substance, confidentiality rests on three levers, in that order of return. The contractual lever: we train exclusively on professional versions of the tools (Microsoft Copilot and Claude, for example), with a data processing agreement, zero data retention and no training on your data, never on free consumer versions. That single switch settles most of the risk. The technical lever: data hosted in Switzerland or the EU where available, single sign-on, fine-grained access rights, the classic trap being to roll out a paid tool without first checking who has access to what in your document systems. The human lever: we help you classify your files by level of sensitivity, so that the minority of genuinely critical data gets reinforced treatment, and draft a short AI usage charter that sets those rules down in black and white. On the Cloud Act: the risk is real but largely mitigated by combining a data processing agreement with zero data retention. If sovereignty is imperative for part of your caseload, the answer is not all or nothing but routing between a local or Swiss tool for routine work and the cloud for complex work.
A model is reliable exactly where you know how to verify. So the question is not whether AI is reliable, but what an undetected error costs on this particular file. In practice, we systematically build three rules into every document summary: the AI must flag missing information rather than invent it, quote the exact extract that supports each statement, and never draw on knowledge outside the document provided. We train your teams to apply this protocol by reflex, not case by case. Human control of every reference, every citation and every date is not negotiable before a text goes into a deed, an opinion or a pleading: Maijin is a training organisation, not a law firm, we issue no legal advice and professional responsibility for your output remains entirely yours.
No text forbids using these tools. What is regulated is what you put into them and what you do with the output. On the Swiss side, there is no general AI legislation. The foundation remains the revised Data Protection Act, in force since September 2023. Switzerland signed the Council of Europe convention on AI in March 2025, and the Federal Department of Justice and Police is due to deliver a preliminary draft for consultation by the end of 2026, with a sector-based rather than cross-cutting orientation. On the EU side, the Digital Omnibus, Regulation (EU) 2026/1744, entered into force on 27 July 2026. It postpones the high-risk obligations of Annex III from 2 August 2026 to 2 December 2027, and those of Annex I to 2 August 2028. Beware the misreading: the general application date of 2 August 2026 is unchanged, as are the prohibitions in Article 5 and the transparency obligations in Article 50. For a firm or a legal department in French-speaking Switzerland, two concrete points to watch: the AI Act applies as soon as a tool's output is used in the EU, and the GDPR applies as soon as there is data on EU residents in the file, which also covers cross-border commuters and French clients. We train your teams to spot these situations and to document the applicable framework; the final compliance assessment remains the responsibility of your firm or legal department.
We are not the vendor of yet another legal tool: we train on what you already have, or on what you are about to choose. In practice, depending on your constraints: Claude under a contract compliant with the Swiss data protection act, ChatGPT and ChatGPT Work, Microsoft Copilot when your organisation is already on Microsoft 365, and your internal tools where you have them. For summarising bulky files with verifiable citations, NotebookLM. For monitoring, the deep research features of ChatGPT or Gemini, and Perplexity. The principle that guides the choice, and that we teach, is the rule of three: one general conversational tool, one research and synthesis tool to digest large volumes, one specialised tool for the role. Productivity rises as you go from one tool to three, and it collapses at the fourth, when nobody masters anything any more and there are four data processing agreements to check. The programme is built during the scoping call around your tools, not around an imposed stack.
A day of training costs CHF 3,000 to 3,500 excluding VAT, travel costs included. A half day is CHF 2,000. A targeted 90-minute workshop is CHF 1,000, and a day of targeted workshops in 90-minute slots is CHF 2,500, with a level 1 training as a prerequisite. The spaced half-day programme is quoted on the half-day rate (CHF 2,000 per session). Two points that matter for a small practice: the rate is per group and not per participant, so training four lawyers costs the same as training twelve and allows markedly deeper tailoring, and there is no minimum number of participants. Only the maximum is fixed, at twelve people per room, because that is what keeps the session interactive. The preparatory 30 to 45-minute scoping call is not charged.
Three common formats. The discovery training, to give the whole team a shared foundation and the first use cases that actually run on their own files. The spaced half-day programme, half-days 1 to 3 weeks apart, our recommended format, which goes from the foundation through to the moves of the role, drafting, summarising, monitoring and governance, with exercises on your real files between sessions. And the highly specific workshop of 90 minutes, one tool for one task: Claude to spot risky clauses in a supplier contract, for instance. On top of that sits the day of targeted workshops in 90-minute slots, which assumes a level 1 training has already been completed. Sessions take place at your offices or remotely, and we adjust the pace to the size of the team and to your diary constraints. The right format is settled in a 30 to 45-minute scoping call, at no charge.
French and English, both fluently. English is not a token option in Geneva: international organisations regularly require the whole session to be delivered in that language. German is available. On a recent engagement we delivered the same foundation twice, in German in Zurich and in French in Geneva, so that both sites started from the same level of understanding. For a legal department or a firm working across both language regions, it is a format we know how to run. Course materials can be supplied in a language other than the one used to deliver the session.
Yes. Between the sessions of a programme, your teams practise on their real files, anonymised where needed, and come back with what got stuck: we correct the prompts and adjust the content to your actual cases. It is that practice between sessions that turns a demonstration into a reflex, not the first session. You leave with the materials in PDF, a prompt library tailored to your tasks and summary sheets, and organisations that want to go further move on to days of targeted workshops or to a governance strand: classifying data by level of sensitivity, drafting a usage charter short enough to be read, a one-page quick-reference sheet. That is the path taken by a network of community health and social care centres in a French-speaking Swiss canton, whose staff charter went from 7 pages to 3 before being shared with several other centres in the canton.
We do not deliver an academic curriculum and we are not the vendor of yet another legal tool: we train on what you already have, or on what you are about to choose. The difference lies in the method, which puts confidentiality and source verification ahead of speed, and in the format: 12 participants maximum, your own files rather than a standard deck, trainers who use AI every day on their own assignments, and a programme built on the pain points identified during scoping. Maijin is a training organisation, not a law firm: we issue no legal advice and professional responsibility for your output remains yours. That framing is what explains the 5.0/5 rating from 51 Google reviews and the documented results of our engagements, from legal monitoring tooled up for a Geneva foundation in the financial sector to AI governance for a network of community health and social care centres.
Yes, on drafting, document summarising, legal and regulatory monitoring, and the preparation of contentious files. It replaces neither legal reasoning nor the decision, both of which stay human. We train your teams to use it as a fast colleague whose work must always be checked, never as a source of truth.
Drafting first versions of documents (contracts, formal notices, opinions), summarising bulky files and regulatory monitoring. Those are the three uses where the gap between time spent without AI and time spent with well-framed AI is most visible, from the very first session.
No. It speeds up research, drafting and summarising, but legal analysis, litigation strategy and the final decision stay human. We train your teams to lean on AI without ever delegating to it what engages their professional responsibility.
Tailored quote
Train your teams in AI
Get in touch with Jean-Baptiste and Rémi, they will be glad to answer your questions
- A 30 to 45-minute scoping call, at no charge
- A programme built around your three priority pain points
- Your confidentiality constraints settled before the first session