AI is a toolbox I use openly, within clear bounds.

Every document I create, collect, or receive is assigned a classification level.

– definitions & examples

– application rules

If your organisation uses a different scale, or none at all, I set the classification at the start of the engagement and submit it for your approval.

An uplift rule applies: a document classified ‘internal’ at your end is treated as ‘confidential’ at mine. Information that circulates freely within your organisation cannot circulate freely at a third party. The other levels remain unchanged.

I store your public, internal, and confidential documents exclusively on Swiss territory, either with providers subject solely to Swiss law, or on my own infrastructure with full-disk encryption.

– special provision

The ‘secret’ level entails substantial operational constraints: no document leaves your infrastructure, access-rights management remains your sole responsibility, and communications slow down considerably. Costs and timelines rise accordingly. As a matter of principle, this level is applied sparingly, only where the nature of the information requires it. In practice, information your organisation routinely shares by email does not qualify, whatever its subject.

Your secret documents therefore remain within your perimeter: they do not reach my infrastructure, even encrypted. You are responsible for providing me with whatever access mechanism you consider appropriate (physical or virtual workstation, secure remote access, or any other arrangement under your control). My notes, extracts and outputs relating to the matter remain within that same perimeter.

– data breach handling

In the event of a suspected breach involving data at any level other than ‘public’, unauthorised access, or any other security incident, I undertake to notify you immediately, within 24 hours at the latest. I take initial containment measures right away to limit the impact, and cooperate fully with you on incident analysis and on any required notifications to the authorities (Federal Data Protection and Information Commissioner, FDPIC).

I do not use an AI tool simply because it is convenient, fast, or fashionable.

I distinguish four AI processing models.

Your classification describes the status of the data within your organization. My processing rules describe what I am authorized to do with it as a third party.

By default, I raise the level of protection when data leaves your perimeter: data that normally circulates within your organization must not circulate freely within mine.

Default permitted models. Depending on the nature of the engagement, your contractual commitments, the presence of sensitive personal data, or the context of the matter, the most restrictive option may apply.

AI may help me analyze, rephrase, structure, compare, synthesize, or produce variants. It never decides what is right, acceptable, or deliverable.

• I review, correct, and take responsibility for the deliverables I provide to you. The use of an AI tool does not reduce that responsibility.
• At the start of the engagement, I inform you of the intended AI uses and the categories of data concerned; I notify you of any changes along the way.
• I maintain a register of the tools used, the tasks concerned, and the categories of data processed, and I include it as an appendix to each deliverable.
• When a deliverable includes AI-generated output that has not been reworked, such as a machine translation or a raw summary, I identify it as such.

An engagement produces two things: a method that can be reused elsewhere, and content that must not be.

– The Principle

I may reuse an approach, an analysis structure, a presentation model, a visual identity, or generation scripts. I do not reuse one client’s content to serve another.

My materials are designed to separate form from content: templates, visual components, and generation scripts on one side; engagement-specific data on the other. When I produce a presentation, client data is injected at document generation time, from sources dedicated to the engagement.

This architecture avoids copied-and-pasted presentations, remnants from old files, unintended traces of other clients, and questionable reuse of materials.

– AFTER THE ENGAGEMENT

At the end of the engagement, the working sources you have provided to me ( files, exports, temporary access, interim documents ) are deleted, unless you request otherwise or a specific need has been agreed.

Final deliverables classified as public, internal, or confidential are retained for ten years on my infrastructure when they have evidential or accounting value. For the others, this reference period applies by default to ensure traceability, unless a shorter period is agreed in accordance with the minimization principle. These documents remain subject to their initial confidentiality level. Secret-level deliverables are an exception: they remain entirely within your perimeter, in accordance with Section I.

This retention is strictly controlled:

1. I am the only person with access to it.
2. I use it only in the context of the engagements you entrust to me.
3. I do not reuse it for other clients.
4. I do not export it to any AI tool.
5. I delete it at your request, unless a contrary legal obligation applies.

Capitalizing on experience does not mean accumulating material for unrestricted reuse. The method improves; your data remains yours.