Manage AI responsibilities
Who can authorise, suspend or modify an AI product, and on the basis of what evidence?
3500 € per session, excl. VAT
In short
Two days to build a governance prototype tailored to a generative AI product: responsibilities, risks, approvals, monitoring and incident management. Participants produce an operational dossier, test their rules through simulation and define a limited pilot, with quality indicators and explicit conditions for suspension.
Why this training?
An AI product evolves with its models, data and instructions. Without explicit decision-making rights, a seemingly minor change can bypass approvals and leave risks without an owner. Governance must link each risk to a verifiable control, a responsible person and a traceable decision. Generative AI helps prepare documents; judgements remain human and context-sensitive. Case study used throughout: The fictitious organisation Calmérys Assistance is preparing an assistant that suggests responses to complaints. An adviser approves each response. A model change causes unjustified refund suggestions: who detects the issue, decides, suspends operation and authorises resumption? Application at work: Have the dossier reviewed by those responsible for product, security and legal matters, then propose a restricted pilot. Collect evidence that the controls work before any decision to expand.
Objectives of this training
- Define the permitted and prohibited uses of the product
- Assign responsibilities for approval, suspension and resumption
- Build a register linking risks, controls and evidence
- Define acceptance thresholds and monitoring rules
- Test the governance and formalise a pilot plan
Who is it for and what are the prerequisites?
Roles and tasks concerned
- AI product managers and digital portfolio managers
- Architects and data, security, compliance and risk managers involved in product decisions
Prior skills
- Have practical experience of generative AI and know its limitations
- Understand the lifecycle of a digital product
- Be able to describe the stakeholders and constraints of a business process
What you will work on
Define the scope of the mandate
Objective: Set the boundaries of the product. Uses, users, autonomy and sensitive decisions.
Map the dependencies
Objective: Identify the elements to be governed. Model, data, instructions, supplier and human approval.
Assess the risks
Objective: Prioritise harmful situations. Errors, disclosure, bias, misuse and obligations to investigate.
Allocate authority
Objective: Assign decision-making rights. Product, business, security and compliance responsibilities; disagreements.
Organise approvals
Objective: Require evidence before authorisation. Reviews, temporary exceptions, acceptance of residual risk.
Assemble the charter
Objective: Make the rules consistent. Traceability, version control and review of decisions.
Define the evaluations
Objective: Measure quality before the pilot. Test set, acceptable responses and justified thresholds.
Build the monitoring framework
Objective: Link alerts to actions. Errors, human corrections, incidents and log minimisation.
Establish change controls
Objective: Prevent uncontrolled changes. Model change, re-evaluation and rollback.
Simulate an incident
Objective: Test the decision-making chain. Assessment, suspension, communication and resumption.
Address shortcomings
Objective: Assess the practical applicability of the rules. Missing controls, ambiguous responsibilities and insufficient evidence.
Prepare the pilot
Objective: Plan a limited trial. Scope, resources, reviews and stopping conditions.
What you take away
- Governance charter and responsibility matrix
- Register of risks, controls and expected evidence
- Approval dossier and monitoring dashboard
- Incident procedure with simulation report
- Pilot plan and decision log
Method and assessment
- Short instructional sessions followed by immediate practice on the running case study.
- Individual work followed by work in pairs, with peer review of the outputs.
- Guided group discussion: what works, what needs correcting and what remains a human decision.
- Preparation: Prepare a fictitious or anonymised description of an AI product, without confidential data Identify the tools authorised by the organisation and the applicable internal policies
Observable criteria: Every critical decision has an owner and an escalation route; Every priority risk has a testable control and expected evidence; Every indicator specifies its calculation, threshold, frequency and owner; The simulation records detection, suspension, adjudication and conditions for resumption; The pilot explicitly sets out its scope, exclusions and stopping criteria.
Frequently asked questions
Is coding required?
No, the prototype focuses on rules, controls and decisions.
Is real data used?
The exercises use only fictitious or anonymised cases and authorised tools.
Is legal compliance established?
No, regulatory issues are identified for validation by the relevant functions.
Is the product deployed during the training?
No, the 14 hours excluding breaks produce a tested dossier and a pilot plan.
Last updated: