From I Don t Know to Trustworthy AI What Leaders Can Learn from Jacques Pommeraud s Big 2025 Talk on Truth

On 23 September 2025, on the Bang stage at Big 2025, Groupe INETUM CEO Jacques Pommeraud took on a deceptively simple theme: "Truth" in the age of artificial intelligence. In just a few minutes, he put three uncomfortable challenges at the center of the AI debate: algorithmic bias, model hallucinations, and malicious misuse.

His message was both disarmingly human and deeply strategic. He encouraged leaders to embrace a healthy skepticism toward AI systems, to rehabilitate the honest "I don’t know", to systematically validate AI outputs against independent evidence, and to stay anchored in their own values when deploying AI at scale.

This is not just philosophical. For companies building their AI roadmap, Pommeraud’s focus on truth translates directly into competitive advantage: more trusted products, more resilient operations, and a stronger license to operate with regulators, customers and society.

Why "Truth" Matters in the Age of AI

AI systems do not seek truth; they optimize objectives. Large language models predict likely next words. Recommendation engines maximize engagement or conversion. Scoring models estimate probabilities. None of these systems have an intrinsic concept of what is true or fair.

For business and public decision makers, that gap matters. When AI output is mistaken for objective reality, three types of risks explode:

• Strategic risk: Leaders take decisions based on plausible but wrong insights generated by models that were never properly validated.
• Operational risk: Teams quietly rely on AI suggestions, workflows or automations that contain subtle but systematic errors.
• Reputational and regulatory risk: Biased or misleading outcomes trigger customer backlash and attract the attention of regulators focused on AI transparency, discrimination and safety.

By putting "Truth" at the center of a high-visibility event like Big 2025, Pommeraud underscored a crucial point: trustworthy AI is not a nice-to-have, it is the foundation of sustainable AI value creation.

The Three AI Truth Challenges Highlighted at Big 2025

The session summary of Jacques Pommeraud’s talk emphasizes three areas every AI leader should be actively managing: algorithmic bias, AI hallucinations and malicious misuse. Each of these has direct implications for corporate AI strategy, risk management and governance.

1. Algorithmic Bias: When History Becomes Destiny

Algorithmic bias occurs when AI systems systematically produce unfair or skewed outcomes for certain groups or situations. This often happens because the data used to train the model reflects historical imbalances, blind spots or discriminatory practices.

Common sources of bias include:

• Skewed training data: Overrepresentation of some populations and underrepresentation of others.
• Proxy variables: Seemingly neutral factors (such as postal code) that correlate with sensitive attributes (such as income level or ethnicity).
• Labeling and annotation bias: Human annotators applying subjective or inconsistent criteria.
• Feedback loops: Biased decisions that feed back into the dataset and reinforce the bias over time.

For companies, the impact is not abstract. Biased models can lead to:

• Discriminatory lending or hiring outcomes and associated legal exposure.
• Unfair pricing or eligibility decisions that erode customer trust.
• Skewed resource allocation, for example in customer service, healthcare or public services.

As regulators around the world tighten rules around non-discrimination, explainability and risk management in AI, bias mitigation is becoming both a moral and a business imperative.

2. AI Hallucinations: Confidently Wrong Machines

Hallucinations occur when generative AI systems confidently produce content that is factually wrong, invented or logically inconsistent. These outputs often sound fluent, authoritative and detailed, which makes them especially dangerous in high-stakes contexts.

Hallucinations are not just rare glitches. They arise from how generative models work: they generate the most likely continuation of a sequence, not the most truthful one. This can produce:

• Invented references, statistics, case law or regulations.
• Fabricated product features or company information in customer-facing content.
• Mismatched or distorted summaries of long documents or datasets.

For SEO, content, legal, finance or healthcare teams, unfiltered hallucinations can directly damage credibility. That is why Pommeraud’s call to validate AI outputs against independent evidence and to embrace the honest "I don’t know" is so strategically important.

3. Malicious Misuse: When Powerful Tools Get Weaponised

Finally, the summary of Pommeraud’s talk points to the danger of malicious use of AI. As models become more capable and accessible, attackers and bad actors can exploit them to:

• Generate highly personalized phishing emails or scam campaigns at scale.
• Create deepfake audio, images or video that can be used for fraud, manipulation or extortion.
• Automate parts of cyberattacks, from vulnerability scanning to code generation.
• Spread convincing misinformation and disinformation at high volume.

Even if your organization never intends to use AI maliciously, you must assume adversaries will. That has direct implications for fraud detection, cyber defense, brand protection and crisis communication strategies.

Rehabilitating "I Don’t Know": A New Leadership Reflex

One of the strongest signals in Jacques Pommeraud’s Big 2025 message is his invitation to rehabilitate the phrase "I don’t know". In a world of always-on AI tools that offer instant answers, admitting uncertainty can feel countercultural. Yet it is exactly what trustworthy AI requires.

In practice, this means leaders and teams should feel empowered to say:

• "I don’t know if this model is fair yet; we have not tested it on diverse populations."
• "I don’t know if this answer is correct; we still need to check it against primary sources."
• "I don’t know if this use case aligns with our values; we need an ethical review."

Far from being a weakness, this reflex is a strength:

• It normalizes rigorous validation instead of blind trust in AI outputs.
• It creates a safe space for experts to challenge the machine and each other.
• It signals integrity to regulators, employees and customers, reinforcing your reputation as a cautious, responsible innovator.

When leaders model this behavior from the top, teams feel permission to treat AI as a powerful assistant, not an infallible authority. That mindset shift is the foundation of trustworthy AI.

A Values-Driven Approach to AI: Putting Ethics Before Efficiency

Pommeraud’s emphasis on relying on one’s own values is a direct challenge to a purely efficiency-driven approach to AI adoption. Cutting response times or increasing conversion rates is not enough if the system undermines fairness, privacy or human dignity.

A values-driven AI strategy typically includes:

• Clear ethical principles (for example, non-discrimination, human oversight, privacy by design, environmental responsibility).
• Governance structures such as ethics councils, AI steering committees or risk review boards that can veto or reshape use cases.
• Escalation paths for employees to flag AI-related concerns without fear of retaliation.
• Transparent communication with customers and partners about how AI is used and what safeguards are in place.

These are not abstract ideals. Organizations that anchor AI in explicit values tend to:

• Accelerate regulatory approvals and certifications.
• Win long-term trust from enterprise customers and public authorities.
• Attract talent that is proud of how technology is being used.

Turning Skepticism into a Strategy: Practical Steps for Corporate AI

How do you translate these ideas from a keynote stage into day-to-day decisions? Below is a practical roadmap to operationalize truth, transparency and ethics in your AI strategy.

1. Start with Clear, Value-Anchored Use Cases

Instead of deploying AI because it is fashionable, start with use cases where:

• The business problem is clearly defined (for example, reducing fraud losses, improving support quality, optimizing logistics).
• The potential impact of errors is well understood (low, medium, high risk).
• The use case can be explicitly mapped to your ethical principles and legal obligations.

For each candidate project, ask:

• What does "truth" mean in this context? Accuracy? Non-discrimination? Faithful summarization? All of the above?
• What are the worst realistic failures, and who would be harmed?
• Which human roles must remain in the loop to protect against those failures?

2. Build Transparent AI Pipelines

Transparency turns abstract trust into something auditable. It empowers teams, auditors and regulators to understand how an AI system was built, trained, deployed and monitored.

Key practices include:

• Documenting datasets (origin, coverage, known limitations, consent and licensing).
• Keeping versioned model cards that describe intended use, known risks, performance metrics and testing methods.
• Logging key decisions and overrides so that you can reconstruct how an outcome was reached.
• Providing meaningful explanations to affected users, where technically and legally feasible.

The table below links the three challenges highlighted at Big 2025 with concrete transparency practices and business benefits.

3. Design for Bias Mitigation from Day One

Waiting to address bias until after deployment is a recipe for expensive rework and reputational damage. A stronger approach is to build bias mitigation into every phase:

• Data phase: Source diverse, representative data. Check for missing groups or systematically different treatments in historical records.
• Modeling phase: Test different model architectures and features with fairness metrics in mind, not just accuracy or profit.
• Evaluation phase: Report performance across subgroups (for example, age, region, language) and scenarios, not just overall averages.
• Deployment phase: Set guardrails such as thresholds, human approvals, appeal mechanisms and regular bias audits.

Bias mitigation is not about achieving perfection. It is about being able to show regulators, customers and employees that you are systematically identifying, measuring and reducing unfair outcomes.

4. Make Validation a Non-Negotiable Habit

Pommeraud’s call to validate AI outputs against independent evidence is a direct antidote to hallucinations and overconfidence. In practice, this means three layers of control:

• Pre-deployment validation: Benchmark models on realistic, curated test sets. Include edge cases, ambiguous cases and historically sensitive situations.
• Human-in-the-loop review: For high-risk use cases, require expert approval before AI recommendations turn into final decisions.
• Post-deployment monitoring: Track error rates, complaint patterns, manual overrides and drift over time.

For generative AI in particular, teams can implement safeguards such as:

• Automatic retrieval of up-to-date facts from authoritative sources before responding.
• Rules that force the model to say it does not know when confidence is low or data is missing.
• Mandatory human review for any legal, medical, financial or HR-related content.

5. Embed Ethical Guardrails and Governance

Ethical guardrails turn principles into enforceable practice. Strong governance clarifies who is responsible, who can approve, and who is accountable when things go wrong. Key building blocks include:

• An AI governance framework that defines roles and responsibilities across business, data, risk, legal and IT.
• A risk-based classification of AI use cases (for example, low, medium, high, prohibited) with corresponding controls.
• Policies and standards that cover data sourcing, model development, evaluation, deployment and decommissioning.
• Alignment with recognized norms and standards such as emerging AI management system standards and sector-specific guidelines.

When governance is clear, teams know when they can experiment freely and when they must escalate, document and seek approval. That balance between agility and control is crucial for scaling AI safely.

6. Prepare for Malicious Misuse Scenarios

Managing malicious misuse starts with acknowledging that your AI assets can be both targets and tools in an attack. Effective preparation includes:

• Threat modeling: Identify how internal or external actors could misuse your models, data or infrastructure.
• Access control and monitoring: Restrict who can use powerful internal models, log usage patterns, flag anomalies.
• Content safety filters: Implement detection and blocking mechanisms for clearly harmful outputs where technically feasible.
• Incident response playbooks: Plan how to respond if your brand is targeted by AI-generated deepfakes or large-scale misinformation.

By treating misuse as a governance and security problem, not just a technology challenge, organizations can both reduce risk and show regulators that they are acting responsibly.

What This Means for SEO, Content and Marketing Teams

Pommeraud’s focus on truth, skepticism and values has direct consequences for how SEO, content and marketing teams use AI every day. These functions are often among the first to adopt generative AI at scale, making them frontline actors in the fight against bias and hallucinations.

AI as a Co-Author, Not a Ghostwriter

Generative AI can dramatically accelerate content production: keyword research, outlines, drafts, meta descriptions, translations, localization and more. But speed is useful only if the content is accurate, trustworthy and aligned with your brand values.

Practical guidelines for SEO and content teams include:

• Always keep a human editor in the loop for factual, legal and brand checks.
• Treat AI output as a starting point to be refined, not a final deliverable to be copy-pasted.
• Explicitly instruct AI tools to admit uncertainty and to refrain from inventing data, references or quotes.
• Maintain a list of authoritative sources that writers must consult for verification.

Building E-E-A-T Through AI Transparency

Search engines increasingly reward experience, expertise, authoritativeness and trustworthiness (E-E-A-T). When AI is part of your content workflow, transparency becomes a differentiator. Consider:

• Explaining internally which parts of content processes are AI-assisted and which are fully human.
• Documenting editorial review steps that ensure factual accuracy and fairness.
• Establishing internal style and ethics guides for AI-generated content.

By aligning your SEO strategy with broader AI governance, you not only reduce risk but also signal to users and partners that your content is carefully curated, not blindly automated.

Fighting AI-Generated Misinformation with Better Information

As AI makes it easier to generate low-quality or misleading content at scale, trustworthy brands have a unique opportunity: become visible beacons of verified information in their niches. This means:

• Publishing clear, evidence-based explainers on complex AI topics such as bias mitigation, transparency, governance and ethics.
• Keeping content updated as regulations, standards and best practices evolve.
• Creating resources that help customers understand how your organization uses AI and what protections are in place.

In this landscape, Pommeraud’s emphasis on truth becomes not only a moral stance but also a powerful positioning for content and SEO strategies.

Governance Checklist: Are You Ready for Truthful AI?

To translate the spirit of Jacques Pommeraud’s Big 2025 intervention into concrete next steps, use the following checklist as a quick self-assessment.

• Values and principles– Have you defined and communicated clear ethical principles for AI use across your organization?
• Use case classification– Do you categorize AI projects by risk level, with corresponding controls and approvals?
• Bias mitigation– Do you have documented processes to identify, measure and reduce bias in data and models?
• Validation culture– Are teams explicitly encouraged to say "I don’t know" and to seek independent evidence before trusting AI outputs?
• Transparency artifacts– Do you maintain model cards, dataset documentation, logs and explanations that can be shared with auditors or regulators?
• Human oversight– For high-stakes decisions, is human review clearly defined, trained and resourced?
• Misuse preparedness– Have you identified how your AI assets could be misused and put in place safeguards and response plans?
• Training and awareness– Are non-technical teams (SEO, marketing, HR, operations) trained on AI risks and good practices?

The goal is not to achieve instant perfection, but to show continuous improvement and clear intent. This is exactly the kind of posture that regulators, partners and customers increasingly expect.

Conclusion: Truthful AI as a Competitive Advantage

From the Bang stage at Big 2025, Jacques Pommeraud brought a simple yet demanding message to the center of the AI conversation: if we want AI to be transformative in a positive way, we must put truth, skepticism and values back at the heart of how we design, deploy and use it.

For corporate leaders, policymakers, SEO teams and AI practitioners, the implications are clear:

• Treat "I don’t know" as a strength that opens the door to better validation.
• Recognize bias, hallucinations and misuse as core strategic risks, not side issues for technicians.
• Build governance, transparency and ethics into AI from day one, instead of bolting them on later.

Organizations that follow this path will not only reduce risk. They will build durable trust with their customers, employees, regulators and society. In a market where AI capabilities are rapidly commoditized, that trust may become the most decisive competitive advantage of all.

In short, the lesson from Big 2025 is both demanding and inspiring: the future belongs to those who can harness AI’s power without sacrificing the truth.