← back to the notebook · talk

A responsible company does not simply hand out ChatGPT, Copilot or Gemini accounts. It decides what it protects, what it allows, what it refuses, and how it trains the people who will live with these tools every day.

This is no longer only a technical matter. It has become a matter of governance, because it touches data, vendors, dependencies, sovereignty and legal responsibility. It has also become a matter of CSR, because it touches working conditions, mental load, attention, free will, and employees' ability to distinguish assistance from full delegation.

The talk starts from a double observation. On one side, companies are right to want to capture the opportunities of generative AI. On the other, they would be wrong to believe that spontaneous adoption is enough. When the company does not frame use, use frames itself: through individual habits, vendor promises, business urgency and cognitive shortcuts.

I Govern to remain in control

AI governance is not about producing a PDF charter that nobody reads. It is about creating a living system: rules people can understand, explicit trade-offs, discussion spaces, named responsibilities, and the ability to say no when a use case looks profitable but damages too much around it.

The talk approaches governance as a concrete discipline, at the crossroads of executive leadership, IT, legal, compliance, cybersecurity, business teams and HR.

• mapWhere AI is already being used, officially or not, with which data, which tools, which vendors and which levels of risk.
• classifyDistinguish ordinary, sensitive, critical and forbidden uses according to data, business impact and degree of automation.
• decideChoose tools, models and access rules while accounting for sovereignty, confidentiality, traceability and vendor dependency.
• operateInstall simple rituals: AI referents, use-case committees, feedback loops, incident reviews and regular updates to the internal doctrine.

The goal is not to slow the company down. It is to prevent a weak, fragmented, dependent and legally fragile adoption. A governed AI is an AI whose meaning, tempo and limits remain in the company's hands.

II Train as a responsible company

Employee training on AI cannot stop at a prompt engineering workshop. Learning how to phrase a better request is useful. But it is not enough for a tool that can change our relationship to effort, attention, decision-making, other people and ourselves.

Training, here, means giving employees a culture of use. Knowing when AI genuinely helps. Knowing when it impoverishes thought. Knowing how to verify, cite, challenge, rewrite and take back control. Also knowing how to protect time and spaces where one thinks without assistance.

• autonomyUnderstand the limits of probabilistic answers, hallucinations, bias, rhetorical confidence and the authority effect of conversational interfaces.
• cognitive healthIdentify risks of fatigue, over-solicitation, loss of attention, dependency on instant answers and erosion of judgement.
• real workAdapt use cases to actual jobs instead of imposing a generic instruction to "use AI" on tasks that are not all suited to it.
• spilloverAccept that habits formed at work follow employees home: into personal searches, parenting, news consumption and daily decisions.

This is the core CSR angle. A company that gives access to these technologies without serious training externalizes the human cost of adoption. A company that trains with rigour recognizes that AI is not merely a productivity lever, but a new cognitive environment.

The point is not to prevent employees from using AI. It is to give them the means to use it without handing over their judgement. Governing AI, Training Humans

III What the talk covers

The format is flexible, but the spine remains the same: start from real uses, understand risks without hysteria, then build a responsible adoption path.

• 01Why AI has become a leadership matter. Shadow AI, competitive pressure, business expectations, reputational risks and managerial responsibility.
• 02Sovereignty and dependency. Data, models, clouds, vendors, traceability, contractual clauses, open-source and proprietary choices.
• 03Reasoned usage. Cases where AI truly augments work, cases where it degrades it, and zones where automation must remain under strong human control.
• 04Mental health and cognitive hygiene. Attentional load, productivity pressure, loss of confidence in one's own thought, porous boundaries between work and personal life.
• 05Training for responsibility. Awareness paths, use rules, business workshops, internal referents, spaces for debate and the right to doubt.
• 06Moving to action. The first decisions to take within thirty days so AI governance does not remain a committee topic.

Possible formats · according to the audience

• 60 to 90-minute keynote: awareness, executive framing, launch of a responsible AI initiative.
• 2-hour talk and debate: talk followed by questions with executive teams, IT, HR, CSR, DPOs, business teams or managers.
• 3-hour masterclass: work on your internal use cases, first doctrine rules, prioritisation of risks and opportunities.
• 1 to 2-day training: deeper awareness, business workshops, usage charter, training plan and operational governance.

What participants take away · concrete, not decorative

• A simple grid to distinguish useful, acceptable, sensitive and forbidden uses.
• A shared vocabulary between leadership, tech, HR, legal, CSR and business teams.
• The principles of a living AI charter that people can actually understand and apply.
• A reading of human risks: cognitive overload, dependency, loss of critical thinking, social pressure to adopt.
• A starting point for building a responsible employee training path.

IV Book this intervention

This talk is for organisations that want to avoid two dead ends: refusal on principle, which lets uses develop elsewhere, and enthusiasm without a frame, which turns AI into one more productivity injunction.

It can serve as the starting point for an AI governance initiative, a responsible digital CSR policy, an executive seminar, an employee training programme, or an internal day dedicated to responsible uses of generative AI.

To discuss it, email is simplest: pierre.vannier@flint.sh. Please include context, audience, intended format and your organisation's current AI maturity.