Skip to content

LLMs and GDPR: using generative AI responsibly in the company

Business teams, lawyers, DPOs and managers · 1 day (7 hours) · in-house, on site or remote · 10 participants maximum

At a glance

Duration
1 jour (7 heures)
Format
On site and remote
Group size
10 participants maximum
Level
Beginner
Audience
Business teams already using generative AI tools, with or without a written rule
Prerequisites
No technical or legal prerequisite. The GDPR notions needed for the day are covered in session.
Methods
Short talks, twenty minutes at most, each followed by a workshop on the tools and documents actually used in the team
Assessment
Beforehand: a positioning questionnaire (tools in use, GDPR knowledge, incidents already encountered), which adjusts the weighting of the modules
Lead time
Scheduled during framing, to suit your availability
Pricing
On a quote basis, depending on format, duration and number of participants. The first call is free.
Accessibility and disability
See the dedicated section

A generative AI tool is a recipient like any other: what you submit leaves the company, sometimes to a subcontractor outside the European Union, often without anyone having checked the contract. This day is not about saving time with AI: it is about setting the rules before use becomes widespread. We start from the GDPR as it already applies (personal data, purpose, informing people, subcontracting, transfers) and confront it with the tools your teams actually use. Each module ends with a written rule of caution that can be applied the next day. By the end of the day, the team leaves with a data sorting grid, a register of uses already started, a draft charter and a list of points for your DPO or your lawyer to settle. That is the principle of this training: I give rules of caution and reflexes, I never say that a use is compliant. That judgement belongs to the people who carry the responsibility.

Learning objectives

  • Recognise personal data, sensitive data and confidential data in a working document before submitting it to a generative AI tool
  • Tell apart a consumer AI tool, a business account and an internal deployment, and identify what each set-up changes for the data passing through it
  • Read the clauses of an AI provider that matter for the GDPR: role (controller or processor), retention, reuse for training, location and transfers
  • Describe a generative AI use as a register entry: purpose, data involved, tool, people informed, duration
  • Identify the situations that trigger an obligation to inform people or the involvement of the DPO, and know who to escalate to
  • Draft a generative AI usage charter on three levels of rules, adapted to your department
  • Handle an access or erasure request concerning data that passed through an AI tool, knowing what is possible and what is not
  • Formulate the questions to ask the DPO or the lawyer before approving a new use, rather than deciding alone

Programme

01What an LLM does with your data

1 hour
  • What happens when text is submitted to a generative AI tool: transit, storage, possible reuse for training depending on the provider terms
  • Three very different set-ups: consumer tool with a personal account, business account, model deployed internally
  • Why a model can reproduce information it has seen, and why you cannot make it forget
  • The most frequent leak scenarios, described without naming any company and without invented cases: a customer contract pasted into a consumer tool, health data submitted for a summary, source code submitted for a fix
  • Round table: which tools are actually used in the team, by whom, for what

02The GDPR in five notions, applied to generative AI

1 hour 15
  • Personal data, sensitive data, confidential data outside the GDPR: three regimes, three reflexes
  • Purpose and minimisation: why “because the tool is there” is not a purpose
  • Controller and processor: who is who when using an external AI tool
  • Informing people: what the GDPR requires when data about a customer, a candidate or an employee goes through a tool
  • Transfers outside the European Union: the question to ask before any other, and the answer you rarely get without reading the contract
  • What this day does not settle: the final legal qualification belongs to the DPO or the lawyer

03Sorting data and choosing the tool

1 hour 30
  • A four-category sorting grid: submitted as is, anonymised first, submitted only in a tool under contract, never submitted
  • Anonymising and pseudonymising: the practical difference, the limits, and why a text “with no names” often remains identifying
  • Reading a provider's terms in twenty minutes: the six clauses to find (role, retention, training, location, sub-processors, security)
  • Workshop: each participant sorts ten document extracts from their own field, then compares their sorting with the others
  • Workshop: guided reading of the terms of use of two tools, compared on the six clauses

04Register of uses, charter and people's rights

1 hour 30
  • Why a register of AI uses, separate from the record of processing activities, and what it must contain at a minimum
  • Writing a use entry: purpose, data involved, tool and account used, people informed, retention period, point of contact
  • A three-level usage charter: allowed, allowed under conditions, never; who signs it, who keeps it alive
  • People's rights (access, rectification, erasure, objection): what can be done when the data went through a tool, and what is technically impossible
  • Workshop: each participant writes the register entry for a real use in their team, then the group drafts the skeleton of the charter

05Processors, contracts and transfers

1 hour
  • What a contract with an AI provider must cover for the DPO to be able to give an opinion: processing clauses, commitments on training, location
  • Sub-processors: the tool provider itself uses other providers, and that counts
  • Transfers outside the European Union: the questions to ask, and who to ask, without claiming to answer them yourself
  • A free tool or a personal account: why there is usually no contract at all, and what that implies
  • Workshop: list of questions to pass on to your DPO or lawyer before approving a new tool

06Information duties, incidents and action plan

45 minutes
  • Informing the people concerned: when, through which channel, in which words
  • When data has gone where it should not have: first-hours reflexes, who to tell, what not to do
  • A thirty-day action plan per participant: three rules to apply, one use to document, one point to have settled
  • Validation quiz and handover of the take-away materials

Who it is for

  • Business teams already using generative AI tools, with or without a written rule
  • Lawyers and DPOs who have to frame these uses without knowing how the tools work
  • Managers who must answer their teams on what is allowed and what is not
  • IT or procurement leads asked to choose or contract an AI tool
  • Organisations handling data about customers, patients, employees or citizens

What you take away

  • The four-category data sorting grid, filled in on extracts of your own documents
  • A template for a generative AI register entry, and the entries written in session
  • The skeleton of your team's usage charter, on three levels, for your DPO or management to approve
  • The list of six clauses to look for in an AI provider's terms, on a single page
  • The list of questions to pass on to the DPO or the lawyer before approving a new use
  • A one-page reflex sheet for a data leak through an AI tool: the first hours, who to tell
  • The session materials, in a readable and printable format

Prerequisites

  • No technical or legal prerequisite. The GDPR notions needed for the day are covered in session.
  • Having some idea, even rough, of the generative AI tools used in your team: a preliminary questionnaire is used to list them.
  • Having a workstation with a browser for the terms-of-use reading workshops.

Teaching methods

  • Short talks, twenty minutes at most, each followed by a workshop on the tools and documents actually used in the team
  • A preliminary questionnaire to the client: tools in use, whether there is a DPO, existing internal rules, business sector
  • Guided reading of real terms of use, in session, rather than summaries
  • Examples drawn from publicly described leak scenarios, without naming any company and without invented cases
  • The register and the charter are drafted in session, not handed over as templates at the end of the day
  • Group capped at 10 participants so that every register entry is reviewed
  • When a DPO or an in-house lawyer attends, they co-run modules 4 and 5: they are the ones who will decide afterwards

Assessment of learning

  • Beforehand: a positioning questionnaire (tools in use, GDPR knowledge, incidents already encountered), which adjusts the weighting of the modules
  • During the day: sorting ten document extracts into four categories, corrected and argued in session (module 3)
  • During the day: a register entry written by each participant, reviewed against five observable criteria: purpose written down, data listed, tool and account named, people informed identified, point of contact designated (module 4)
  • At the end: a fifteen-question validation quiz on the five GDPR notions, data sorting and incident reflexes, corrected collectively
  • Four to six weeks later: a one-hour remote follow-up on what was applied from the action plan, on the register and on the answers obtained from the DPO or the lawyer

Where what I teach comes from

Adrien Bouthet, freelance IT Project Manager and Product Owner based in Bordeaux, with eight years of experience steering digital projects. Since February 2025 I have been steering the digital onboarding journeys of La Banque Postale (account opening, minor and adult journeys, Livret A subscription): a budget in the order of one million euros, under constant KYC and GDPR constraints, including raising team awareness of data governance within my scope. In that environment, the question “may this data go into this tool?” is anything but theoretical. Those figures are results from steering assignments: they say where what I teach comes from, they are not a promise made to participants. I have taught this “generative AI and GDPR” course in a school, to students who arrive in companies with the tools already in their pocket, and I have been a trainer in programming and management since 2020. One point that matters: I am neither a lawyer nor a legal adviser. This day passes on rules of caution and a project manager's reflexes; the legal qualification of a use remains the responsibility of your DPO or your counsel.

Real example

The data part of this day comes from a field where it is not negotiable: since February 2025 I have been steering the digital onboarding journeys of La Banque Postale, a budget in the order of one million euros under constant KYC and GDPR constraints, including raising team awareness of data governance. A result from a steering assignment, not a promise made to participants.

Read the case study

Frequently asked questions

Does this day say which AI uses are GDPR compliant?+

No, and that is deliberate. I am neither a lawyer nor a legal adviser: I give rules of caution, a sorting method and the right questions to ask. The legal qualification of a specific use belongs to your DPO or your lawyer, and the day is built to make their work easier, not to replace it.

How is this different from the “Everyday generative AI” training?+

The other day looks for the uses that save time and works on your documents to produce meeting reports or drafts. This one contains no productivity workshop: it focuses on rules, the register, the charter, contracts and people's rights. The two follow each other well, in either order.

There is no DPO in the company. Is that a blocker?+

No. The day pinpoints the questions that fall to a DPO or a lawyer, so they can then be put to an external adviser in one go rather than piecemeal. If appointing a DPO is mandatory in your situation, that is one of the points the question list will surface.

Are technical skills required?+

No. No code is written and no tool is configured. How a language model works is explained in the first module, limited to what matters for data. Lawyers and DPOs find the technical part they lack, business teams the regulatory part.

On site or remote?+

Both. On site at your premises, in Bordeaux and across Nouvelle-Aquitaine; remotely anywhere in France, with the same sorting and drafting workshops. Remotely, the documents used for the sorting exercise are sent in advance and contain no real personal data.

How many participants at most?+

Ten. Each participant writes a register entry that is reviewed in session; beyond ten, that review no longer fits in the day. For a larger headcount, several sessions are scheduled, preferably by department, since uses differ.

Are the documents used during the day kept?+

No. The extracts brought for the sorting exercise remain company property and are neither kept nor reused after the session. Participants are in fact asked to bring extracts already stripped of real personal data: the exercise is about recognising categories, not about handling real data.

Framework and terms

Funding

These sessions are invoiced directly to the company, out of its own skills development budget. They are not eligible for pooled funding schemes (OPCO, CPF, France Travail), which require Qualiopi certification.

Certification

This session is not a certifying course. I do not deliver any Scrum.org, Scrum Alliance, Scaled Agile (SAFe) or PeopleCert (ITIL) certification, nor any qualification registered with the RNCP or the Répertoire spécifique. The goal is operational: practices you can apply, not a diploma.

Contractual framework

Each session is covered by a professional training agreement (convention de formation professionnelle) signed before the start, setting out the objectives, the programme, the duration, the delivery arrangements and the agreed fee. The internal rules applicable to the session are handed to participants at the opening.

Certificate of attendance

A certificate of completion is handed to each participant at the end of the session. It states the objectives, the nature and the duration of the training, along with the results of the assessment of learning.

Accessibility and disability

Disability referent: Adrien Bouthet, contact@adrienbouthet.fr. Any situation requiring an adjustment (access to the premises, adaptation of the materials, of the pace or of the assessment arrangements) is examined during the framing phase, within 5 working days. If the adjustment exceeds our means, we refer to specialised resources (Agefiph, Cap emploi).

Setting the rules before use spreads

Book a free first call. We look at the tools already used in your teams, whether you have a DPO and your sector, and I tell you straight whether this day is the right format or whether your need calls for legal advice first.

The matching service

Generative AI integration