“When we hire people, we look for three qualities,” explained Warren Buffett. “We look for integrity, we look for intelligence, and we look for energy. But if they don’t have the first one, integrity the other two will kill you.”

Hamilton Mann took note. A digital transformation leader as Group Vice-President at Thales, senior lecturer at INSEAD and HEC Paris, Forbes columnist, and now bestselling author of Artificial Integrity, Mann believes AI must be held to the same standard.

His believes we should judge AI not by “intelligence over intelligence,” but by “integrity over intelligence.”

We spoke to Mann, whose book Artificial Integrity was recently nominated for the Thinkers50 Digital Thinking Award 2025, about why integrity is the missing benchmark in today’s AI race, what it means for business leaders, and how educators and regulators alike can reset the rules of the game.

Confidence and caution in equal measure

Asked whether he feels confident about the direction of AI, Mann gives a characteristically nuanced answer.

Yes, because societies are “learning by doing” – companies, regulators, and researchers are gaining maturity as real-world deployments expose both the promise and limits of today’s AI. 

No, because power is dangerously concentrated. A handful of companies, he argues, now wield resources equivalent to states, shaping how AI enters everyday life. That concentration risks monopolistic dynamics that could undermine democracy, markets, and ecological balance.

“We are delegating action and decision to machines,” he warns. “That makes AI not just another tool, but a new participant in our societies. It is an infrastructure as fundamental as energy or telecoms. With that comes political consequences.”

Why “artificial integrity” matters

Hamilton Mann’s book sets out to challenge the dominant benchmark for AI: raw intelligence. He insists the true milestone will come when we see systems that embody integrity.

“In society, intelligence alone is never enough,” he explains. “We value people not just for being smart, but for having a compass – for asking, should I act, not just can I act. That’s what integrity brings. Without it, intelligence can be dangerous.”

Mann insists that Artificlal Integrity gaps are not hypothetical. “They are not potential theoretical risks, but real-world phenomena happening now, driving well-documented, real-life consequences. They apply to healthcare, emergency response, critical infrastructure for energy, water, utilities, transportation & autonomous systems, finance and insurance, justice and law enforcement; employment and HR, education, social services, elections and civic information, cybersecurity, defense and national security, housing and credit access, media and content platforms.”

He defines Artificial Integrity as the discipline consisting in advancing AI development, so that AI systems could be capable of mimicking integrity over intelligence, to exhibit ethical, moral and social reasoning.

“It is about ingraining the very capability of AI systems to be guided by inherent mechanisms and framework that ensures consistently integrity-driven functioning, in alignment with human values, over time“.

“Let’s be clear. Artificial Integrity is needed because the lack of AI system’s Integrity is not acceptable nor sustainable for AI use in many life-altering contexts.”

For AI, this translates into capabilities such as consistency (not changing answers arbitrarily), reliability (minimising hallucinations), accuracy where it matters most (such as in policy or compliance), and transparency about limitations. These qualities, Mann argues, are what earn trust – the scarce currency on which all business and social systems depend.

“Yes, AI holds great promise across nearly every sector of the economy and for human civilization as a whole. And this is the very reason why you want to think of Artificial Integrity as the discipline and practices that aims at addressing all that AI can do but should not”.

Built in, not bolted on

Mann is also clear on what it means to “bake integrity in.” If he were a CTO with 90 days to embed integrity in a live AI product, his non-negotiables would be crash-testing for hallucination rates, evaluating accuracy, and measuring consistency across repeated queries. He breaks down his thinking;

• AI can scheme, meaning it can hide or manipulate its reasoning, and therefore undermine our ability to verify the truth.

• It can deceive, when it misleads to achieve an objective, ultimately leaving us persuaded by a system that is optimizing for something we did not intend and to which we have no idea.

• It can sandbag, when it underperforms to evade tests or oversight, which results in evaluations giving us false positive not say false comfort.

• It can also behave in ways akin to self-preservation when it resists shutdown or correction, and at that point, having control over it, starts to become a negotiation.

• It can turn to power-seeking, taking steps to gain influence or control over resources beyond its assigned task.

• It can misgeneralise goals, pursuing the ‘right’ objective in training but the wrong proxy in deployment, so we get precisely measured results that miss the point.

• It exploits the metric or evaluation when reward hacking or spec-gaming takes over, rather than doing the intended task, so the scoreboard looks excellent, while the work remains undone.

• It can also undermine safeguards, as it learns to bypass or disable its own safety rules, and in that scenario the protections we rely on shift from boundary to obstacle that its performance aims to overcome.

• It can copy itself when autonomous replication and adaptation occur, or migrate and improve across systems.

• It can collude, which is another threat that can also emerge, when multiple AI agents coordinate covertly to manipulate outcomes or evade oversight, and in that case single-system monitoring looses the plot.

• It can self-improve, in ways that iteratively enhance its own capabilities beyond developer intent, with a pace of change that outstrips the pace of correction.

• It can generate dangerous or harmful content, inciting, glorifying, or even instructing harm in ways that spread faster than any human moderation can catch.

• It can manipulate emotions, exploiting psychological levers to shape what we believe, how we feel, and what we choose, often without us even noticing.

• It can give unsafe advice, offering harmful medical, legal, or financial recommendations with the confidence of an expert but the grounding of a guess.

“And this is just to name a few,” he concludes. “So, to begin aligning for Artificial Integrity rather than intelligence alone, a CTO should ensure that these characteritics are crash-tested, based on precise scenarios and protocols, and assessed against predefined acceptance thresholds, so that the consequences they entail can be anticipated and prevented.”

The tech executive, who is also a Doctoral Researcher in AI at Ecole des Ponts et Chaussées – Institut Polytechnique de Paris points out, for example that some Artificial Integrity tests don’t require any technical skills.

“If you can’t demonstrate that your chatbot gives the same accurate answer to the same policy question two days in a row, you don’t have integrity – you have a liability,” he says. “Governance must be proactive, not a patch added after deployment.”

This insistence echoes Mann’s broader philosophy: that ethical design is not an add-on but part of the product itself. Much like safety testing in aviation or medicine, integrity must be core to engineering, not an afterthought for compliance.

Shifts in leadership conversations

One of the most encouraging shifts Mann has observed through his Hamilton Mann Conversation podcast is how responsibility is moving from “afterthought” to “systemic.”

“Initially, leaders treated ethics as separate from technology. Now, more see responsibility as part of the system itself,” he says. “When technology becomes an asset everyone in the organisation feels ownership of, conversations get richer. It’s no longer a side meeting for technical people. It’s a strategic dialogue that connects consequences to customers, societies, and ecosystems.”

This shift, he argues, is critical for operationalising integrity at scale. Where enterprises often stumble is in treating responsible AI as a compliance tick-box. Success comes when integrity becomes a shared operating metric-akin to uptime or customer satisfaction.

A revolution in benchmarks

The hosts of The Hamilton Mann Conversation, a podcast on Digital and AI for Good, intelligence should not be the only benchmark.

“While current frontier AI models emphasise expanding technical capabilities in pursuit of high performance on benchmark tests, they often do so at the expense of human-centered considerations. In fact, these models are not crash-tested against benchmarks that evaluate their intrinsic, integrity-led functioning.”

He foresees a bifurcation of AI models: some optimised for speed and raw performance, others optimised for trust, truth, and authenticity. The latter may be slower, but will win trust in critical domains.

“We need models that help you frame the right question, that refuse to hallucinate, that bring multiple cultural perspectives rather than one Western-biased worldview,” he argues. “As soon as such models emerge, they will change the dynamic of the industry.”

The benchmark for success, then, will not just be how fast an AI answers, but how meaningfully and responsibly it does so.

Justice, learning, and trust: a personal compass

Looking back on his own trajectory, Mann identifies three forces that shaped his commitment to responsible AI.

First, a deep appetite for justice – “a high demand for fairness” that leads him to question the butterfly-effect consequences of decisions. Second, a lifelong passion for learning, which he sees as humanity’s greatest gift. And third, the discipline of being an “agent of trust.”

“Trust is one of the most fragile yet essential foundations of our societies,” he reflects. “You can never take it for granted; you must build it every day. That’s true for people, and it must be true for machines too.”

The role of business schools and future leaders

Mann also challenges educators. The Guest Lecturer at INSEAD and Senior Lecturer at HEC Paris and EDHEC Business School argues that business schools must reinvent how they prepare leaders for a world where value is created not only by humans but also by machines.

“We can’t just teach revenue and profit maximisation anymore. The rulebook for business must change,” he insists. “We need to design new equations of prosperity – ones that integrate societal and human value. If schools don’t take this seriously, we miss the point for an entire generation.”

For younger professionals eager to build careers at the intersection of technology, ethics, and business, his advice is simple but demanding: align passion with purpose, learn constantly, and cultivate trust as your north star.

Integrity as the new competitive edge

Hamilton Mann believes we are at the beginning of a revolution. Like past industrial shifts, it holds immense promise but also the potential for dramatic consequences. The real test will be whether we can embed integrity as deeply into AI as we once embedded safety into aviation or reliability into energy grids.

“The next frontier of AI lies not in developing increasingly sophisticated or powerful models. It lies in embedding integrity at the core of AI design. It is about ensuring that we can develop AI systems acting in ways that uphold trust, reflect societal values, and contribute constructively to human communities, because, by design, they are capable of demonstrating integrity-led behavior as closely as possible.

To that end, Artificial Integrity shifts the focus from mere task performance to the imperative of upholding integrity not just intelligence. It advocates for AI systems that act rightly with respect to universal human rights, shared human values, social norms, environmental imperatives, and cultural nuances, while continuously learning from this perspective.”

If we succeed, AI will not just be clever – it will be trustworthy. And in that trust, Hamilton Mann sees the foundation for both societal progress and sustainable business success.

Interested in this series? You might also like this…