Responsible AI in Practice: Insights from Jacques Pommeraud at the Cercle de Giverny

Responsible AI has moved from academic debate to boardroom priority. In an in-depth interview at the Cercle de Giverny, Jacques Pommeraud explores how organizations can turn ethical intentions into operational reality. His discussion connects high-level principles like transparency, fairness, privacy and accountability with the real constraints of regulation, data, technology and organizational culture.

This article draws on the themes and ideas highlighted in that conversation, translating them into a practical roadmap for executives, legal teams, data leaders and practitioners who want to implement responsible AI and communicate trustworthiness to users and customers.

Why Responsible AI Is Now a Strategic Imperative

Jacques Pommeraud situates responsible AI in a moment of rapid change: AI capabilities are accelerating, regulations are tightening and public expectations around ethics and trust are rising. Responsible AI is no longer a “nice to have” – it is a strategic differentiator.

• Customers expect clarity on how AI makes decisions that affect them.
• Regulators are introducing new standards and enforcement powers.
• Employees want to work for organizations that use AI in line with their values.
• Investors and partners increasingly assess AI risks as part of due diligence.

In this context, responsible AI is about much more than risk containment. Done well, it becomes a source of competitive advantage: enabling faster adoption, smoother compliance, stronger customer loyalty and better long-term innovation.

Four Core Principles of Responsible AI

Throughout his conversation, Pommeraud anchors responsible AI in four recurring pillars: transparency, fairness, privacy and accountability. These principles echo leading regulatory and ethical frameworks, but he focuses on how to make them actionable.

Transparency: Making AI Understandable and Explainable

Transparency means that people affected by AI can meaningfully understand how it works and what it does. This is not about revealing source code; it is about offering clear, accessible explanations aligned with the needs of each audience.

• Users and customers need to know when they are interacting with AI, what data is used and what it means for them.
• Business stakeholders need to understand the model’s purpose, limitations and performance characteristics.
• Regulators, auditors and legal teams need documentation of design choices, data lineage, risk assessments and controls.

Pommeraud’s emphasis on transparency translates into practical steps such as model cards, plain-language summaries, clear labelling of AI-generated content and internal documentation that survives staff turnover and scale.

Fairness: Tackling Bias and Unequal Outcomes

Fairness is about recognizing and reducing unjust or discriminatory outcomes, especially for vulnerable or historically marginalized groups. AI systems can reproduce and amplify existing social biases if they are trained on imbalanced or biased data, or if teams are not intentional about equity from the outset.

In practice, fairness work means:

• Defining fairness for each use case (e.g. equal opportunity, equal error rates, no unjustified disparate impact).
• Testing models for performance across demographic or relevant subgroups, not just on average.
• Interpreting trade-offs transparently when fairness metrics conflict with each other or with pure accuracy.
• Building diverse teams and involving affected stakeholders in design and evaluation.

Instead of treating fairness as an afterthought, Pommeraud positions it as a design constraint – something to consider from problem framing all the way through deployment and monitoring.

Privacy: Respecting Data as a Human Right and Strategic Asset

Privacy is more than legal compliance; it is a fundamental expectation from users and an essential component of trust. AI systems often rely on large volumes of data, which intensifies the need for robust data governance and careful risk management.

Key privacy practices that align with the interview’s themes include:

• Data minimization– collecting and retaining only what is necessary for clearly defined purposes.
• Purpose limitation– avoiding repurposing data for new uses without proper legal basis, impact analysis and user awareness.
• Security by design– embedding encryption, access controls, monitoring and incident response into AI pipelines.
• Privacy-preserving techniques– such as anonymization, pseudonymization and, where appropriate, techniques like differential privacy or federated learning.

By treating privacy as a design parameter rather than a late-stage hurdle, organizations can innovate faster while facing fewer legal, reputational and operational risks.

Accountability: Clear Responsibility, Not “The Algorithm Did It”

Accountability means there is always a clearly identifiable human or organizational owner for AI systems and their outcomes. It is the antidote to the temptation of blaming “the algorithm” for decisions that impact people’s lives.

Accountability in responsible AI typically involves:

• Assigning named owners for each AI system, covering design, deployment, monitoring and decommissioning.
• Establishing RACI models (Responsible, Accountable, Consulted, Informed) for cross-functional governance.
• Defining escalation paths when issues arise, including for ethics or rights-based concerns.
• Keeping auditable records of decisions, approvals and changes to models or data.

Pommeraud’s framing encourages leaders to treat accountability as a management discipline: clear roles, documented processes and governance structures that can stand up to scrutiny.

From Ethics to Governance: Operationalizing Responsible AI

One of the most valuable aspects of Pommeraud’s intervention is the link between ethical intent and concrete governance. Principles alone do not change outcomes; organizations need processes, tools and culture to make responsible AI real.

Bias Mitigation as a Continuous Lifecycle

Bias mitigation is not a single technical step; it is a lifecycle that spans problem framing, data collection, model development and post-deployment monitoring.

• Upfront scoping– questioning whether an AI approach is suitable, what harms could occur and who might be affected.
• Data review– checking for missing groups, historical biases or label quality issues.
• Model experimentation– comparing models not only on accuracy but also on fairness and robustness metrics.
• Testing and validation– running scenario tests, edge cases and subgroup analyses before launch.
• Monitoring in production– tracking drift, performance changes and emerging bias over time.

A lifecycle view aligns with Pommeraud’s broader message: responsible AI is a process, not a one-off project or checklist.

Data Governance as the Foundation of Trustworthy AI

Robust data governance is indispensable for responsible AI. Without clarity on data sources, quality, lineage and permissions, organizations simply cannot guarantee responsible behavior from their models.

Effective AI-oriented data governance typically includes:

• Data inventories and catalogs– knowing what data exists, where it resides and who owns it internally.
• Access management– ensuring that only authorized personnel and systems can use sensitive data.
• Quality controls– putting in place validation checks, de-duplication, standardization and issue tracking.
• Retention and deletion rules– aligned with regulatory requirements and user expectations.

Pommeraud’s focus on governance reflects a clear lesson: responsible AI is impossible without a strong data backbone. Organizations that invest here can build AI at scale with more confidence and fewer surprises.

Human-Centered Design and Human-in-the-Loop Oversight

Responsible AI is ultimately about people, not technology. Pommeraud highlights the need for human-centered design and meaningful human-in-the-loop oversight, especially in high-stakes contexts such as health, employment, finance or public services.

Human-centered AI design asks questions like:

• Who is the primary user, and what problem are we really solving for them?
• What information, controls or choices do users need to feel in command, not overruled?
• How do we design interfaces that communicate uncertainty, limitations and risk?
• When should humans have authority to override, review or contest algorithmic decisions?

Instead of fully automating every step, human-in-the-loop approaches keep expert judgment at critical decision points. This reduces risk, increases acceptance and creates a healthier partnership between humans and machines.

Stakeholder Engagement and Ethical Deliberation

Another recurring dimension of responsible AI is stakeholder engagement. Technical teams alone cannot anticipate every ethical concern or social consequence. Pommeraud’s governance-focused perspective encourages organizations to broaden who is at the table.

Depending on the use case, stakeholders may include:

• Internal experts (legal, compliance, risk, HR, security, operations).
• Frontline staff who will use or be affected by AI tools.
• Customers, user representatives or community groups.
• External advisors, ethicists or domain specialists.

Structured engagement – for example, through ethics committees, advisory boards, consultations or co-design workshops – helps surface blind spots early and builds legitimacy around AI initiatives.

Navigating the Evolving Regulatory Landscape

Responsible AI and regulation are converging quickly. Pommeraud situates AI ethics within a growing framework of hard law and soft law: from data protection rules to forthcoming AI-specific regulations.

Key Regulatory Themes Organizations Must Anticipate

Across jurisdictions, several recurring regulatory expectations are emerging:

• Risk-based classification– stricter controls for high-risk uses (e.g. safety-critical or rights-impacting applications).
• Data protection and security– alignment of AI systems with privacy regulations and cybersecurity standards.
• Transparency and user information– obligations to inform individuals when AI is used and how decisions are made.
• Human oversight– requirements to ensure that AI does not entirely replace human judgment in sensitive decisions.
• Documentation and traceability– expectations for detailed technical and organizational documentation.

Instead of viewing these as constraints, Pommeraud frames them as opportunities to strengthen trust and discipline across the AI lifecycle.

Compliance by Design Versus Last-Minute Fixes

A recurring message is the value of compliance by design. Building AI systems with regulatory requirements in mind from day one is more effective and cost-efficient than retrofitting compliance right before launch.

Compliance by design typically includes:

• Embedding legal and risk experts in product and data teams early.
• Using standardized impact assessments for privacy, ethics and human rights.
• Documenting design decisions as you go, not only at the end.
• Setting default technical and organizational safeguards aligned with the strictest plausible standards.

Pommeraud’s approach aligns responsible AI with robust risk management: anticipate, design, document and iterate, rather than wait for an audit or incident to expose gaps.

Practical Governance: Roles, Responsibilities and Structures

Turning responsible AI from aspiration to reality requires clarity about who does what. Pommeraud’s governance-driven perspective encourages organizations to move beyond informal discussions to structured responsibilities.

Example Role Matrix for Responsible AI

The table below shows a simplified example of how responsibilities can be distributed across key functions. Each organization will adapt it to its size, sector and risk profile.

While structures vary, Pommeraud underscores a simple idea: responsible AI needs named owners, clear mandates and cross-functional collaboration; otherwise, good intentions get lost in organizational complexity.

A Roadmap for Organizations Getting Started

For organizations at early or intermediate stages, Pommeraud’s perspective can be distilled into a practical roadmap that balances ambition with pragmatism.

1. Clarify Your AI Ambition and Risk Appetite

Start by aligning leadership on why you are adopting AI and what level of risk is acceptable. Key questions include:

• What strategic value do we expect from AI (efficiency, growth, new services)?
• Which domains are off-limits or require stricter oversight?
• How much uncertainty are we prepared to accept in high-impact decisions?

A shared understanding at the top makes it easier to design coherent governance, policies and communication.

2. Establish Principles, Policies and an AI Governance Body

Translate values into written, actionable guidance:

• Define a concise set of AI principles based on transparency, fairness, privacy and accountability.
• Draft policies that explain what these principles mean for data use, model development, vendor selection and deployment.
• Create an AI or ethics committee (or extend existing risk committees) to oversee sensitive use cases and escalate complex issues.

Pommeraud’s focus on governance reminds organizations that clear guardrails actually accelerate innovation by reducing uncertainty and internal friction.

3. Map Your AI Use Cases and Risks

Before investing heavily, take stock of where AI is already used and where it is planned. For each use case, evaluate:

• Business objective and expected benefit.
• Potential impact on individuals or communities.
• Sensitivity of data involved.
• Regulatory exposure and reputational risk.

This portfolio view helps prioritize efforts: high-impact, high-risk use cases should receive deeper scrutiny, more robust controls and stronger explainability.

4. Integrate Responsible AI into the Development Lifecycle

Responsible AI should be embedded into existing development workflows, not bolted on at the last minute.

• Discovery and design– include ethics, legal and domain experts early; identify affected groups; consider alternative approaches.
• Data and modelling– enforce data quality standards; document sources; run bias and robustness checks; track experiments.
• Testing– validate with real or representative users; test edge cases; evaluate explainability and usability of outputs.
• Deployment– implement monitoring, alerts and rollback mechanisms; log decisions and model versions.
• Operations– monitor for drift, incidents and user complaints; schedule periodic reviews and re-certifications.

This lifecycle approach is fully aligned with Pommeraud’s message: responsible AI is a living process, maintained over time, not a single milestone.

5. Build Skills, Awareness and Culture

Finally, responsible AI requires people who understand both opportunities and risks. Organizations can invest in:

• Training for data scientists on ethics, legal constraints and human-centered design.
• Awareness programs for executives and managers on AI capabilities and limitations.
• Practical playbooks, templates and checklists that translate abstract principles into daily practice.

By normalizing conversations about trade-offs, impacts and safeguards, organizations create a culture where responsible AI becomes the default, not the exception.

Communicating Trustworthiness to Users and Customers

Pommeraud also stresses that responsible AI is not only about internal controls; it is about how organizations show up to the outside world. Transparent, honest communication builds resilience and brand value.

Elements of Effective AI Communication

To earn and maintain trust, organizations can focus on:

• Clarity– plain-language explanations of where and why AI is used.
• Choice– options for human assistance, alternative channels or opt-outs where feasible.
• Control– ways for users to correct data, contest decisions or request human review.
• Consistency– aligning public messages with actual practices, including during incidents.

In line with Pommeraud’s insights, organizations that communicate thoughtfully about AI not only reduce misunderstandings, they also differentiate themselves as reliable partners in an increasingly automated world.

Key Takeaways for Practitioners, Legal Teams and Executives

Jacques Pommeraud’s intervention at the Cercle de Giverny offers a strong, practice-oriented perspective on responsible AI. Bringing his main messages together, several takeaways emerge for different audiences.

For Executives

• Treat responsible AI as a strategic lever, not a constraint.
• Set clear principles, risk appetite and expectations for your teams.
• Invest in governance, data foundations and skills early – they pay off in speed and resilience.

For Legal, Risk and Compliance Teams

• Move from reactive review to proactive partnership in AI initiatives.
• Develop simple, repeatable processes for impact assessments and documentation.
• Translate evolving regulations into practical, technology-aware guidance.

For Data, Product and Engineering Teams

• Embed fairness, transparency, privacy and accountability into daily workflows.
• Document decisions and assumptions; future you (and your auditors) will need them.
• Collaborate with non-technical stakeholders early to anticipate real-world impacts.

Conclusion: Responsible AI as a Catalyst for Better Innovation

Responsible AI is often framed as a set of restrictions. Jacques Pommeraud’s perspective at the Cercle de Giverny offers a more constructive view: ethics, governance and regulation can actually enable more sustainable, scalable innovation.

By grounding AI initiatives in transparency, fairness, privacy and accountability – and by backing those principles with concrete governance, data discipline and human-centered design – organizations can unlock the full value of AI with confidence. In doing so, they not only manage risks, but also build deeper trust with users, customers, regulators and society at large.

For leaders who want to harness AI responsibly, the path is clear: start with principles, turn them into processes, and keep people at the center of every decision. That is the heart of responsible AI in practice.