The case for AI integrity
Is the race to get into AI leaving ethics behind? Hamilton Mann argues that if you sacrifice morals for speed, you might not win in the long term. Even worse, you could end up deferring some costly consequences instead.
Where do ethics fit in the fast-growing world of AI and business? Hamilton Mann has a plan to address that question. He argues that letting ethics slip is not an inevitability; it is a choice. In his book, he argues that “artificial integrity” – or designing AI systems that are genuinely aligned with human values rather than optimized for performance – is not a constraint on innovation, but a condition for it. In an engaging and wide-ranging Zoom conversation, he outlines how organizations can build AI that serves people rather than replaces them, and why, in the long run, integrity could be a vital component and lasting competitive advantage.
You say that the aim of artificial integrity is to anchor the role of AI in acting as a partner to humans and to facilitate work and life in a way that is aligned with values and grounded with integrity. But in practice, whose values should AI be aligned with? And how do we resolve conflicts when the values of individuals diverge with organizational and societal values?
That’s a very profound question. Artificial integrity doesn’t assume a single set of values. It assumes plural values and therefore a need for a legible way to arbitrate among them. So the question isn’t “whose values” in the abstract. It’s very much about which values apply to a specific decision context and by what legitimate process. So you can think of it as three layers in terms of hierarchy: The first layer is the nonnegotiable baseline. It covers societal, legal and human rights constraints and imperatives. And this is where societal values take precedence over organizational or individual values. AI should never violate fundamental rights, safety, antidiscrimination principles. This is the red line layer.
The second layer is the organizational values – whether it is about the mission, the duty of care, the purpose, the ethical stance that you can put in place. It is a way to put together what is mattering from an organizational perspective. This is also a nonnegotiable second line which covers who we are in terms of organization, what we care about, and especially what we care about before and beyond profit and loss. This is the kind of glue that gets a lot of people working in the same organization to be aligned on a way of doing things and the way of behaving in doing those things. So this is what I’m calling the purple line layer.
Then you have the final third layer, which is the individual values. It is where AI should be adapting, should be able to personalize the interaction, frame outputs in a way that is fitting the individual interest. But personal values should only shape experience and individual choices, but not override the red line nor the purple line.
So when values diverge between those three layers, AI systems should be designed to follow a practical conflict resolution method. It is a kind of integrity arbitration algorithm considering two dimensions: the level of stakes and the level of reversibility. If it is low stakes and fully reversible, then the algorithm should prioritize individual values. If it is medium stakes, with potential impact on the red and purple lines, the AI system should apply guardrails before prioritizing individual values. And if this is high stakes, coupled with non-reversible action, then the algorithm should prioritize organizational and societal common interest. Also over time, because such things are not a fixed target – values may evolve as societal feelings evolve.
So when values diverge between those three layers, AI systems should be designed to follow a practical conflict resolution method. It is a kind of integrity arbitration algorithm considering two dimensions: the level of stakes and the level of reversibility. If it is low stakes and fully reversible, then the algorithm should prioritize individual values. If it is medium stakes, with potential impact on the red and purple lines, the AI system should apply guardrails before prioritizing individual values. And if this is high stakes, coupled with non-reversible action, then the algorithm should prioritize organizational and societal common interest. Also over time, because such things are not a fixed target – values may evolve as societal feelings evolve.
Multiple values, one system
Hamilton Mann, author of Artificial Integrity, argues that AI must hold several layers of values at once – societal, organizational and individual – without losing its bearings.
“International frameworks only become real through the daily decisions of individuals acting inside those institutions.”
You say: “Let’s not ask what AI can do for each of us, let’s ask ourselves what we can do to ensure AI serves humanity.” Can that really be achieved on an individual or company level? Doesn’t that require a coordinated international approach?
The short answer is that individual action is necessary but not sufficient. AI is too powerful, too interconnected and too economically consequential to be governed meaningfully by isolated personal goodwill alone. International coordination is indispensable, especially on safety standards, compute governance, cross-border accountability and the protection of fundamental rights. Let’s take the example of nuclear. States do not manage nuclear safety only as a domestic matter. They rely on international conventions and the International Atomic Energy Agency to establish common safety principles. Or the International Civil Aviation Organization, pharmaceuticals, banking, automotive. For any capability used globally, shared guardrails are required.
But the opposite mistake would be to conclude that because systemic coordination is needed, the individual level is irrelevant. It’s not. International frameworks only become real through the daily decisions of individuals acting inside those institutions. I do not believe the individual level is enough on its own, but I also do not believe international coordination can succeed if individuals just abdicate responsibility and wait for a perfect global framework. Any global shared rules come as a result of a movement that started at the individual level at some point, in some place.
Their initial reaction was: “Well, sure, yeah, it’s obvious. Everybody knows that. Yeah, you feel great when you make progress in your work. Sure.” And it was kind of dismissive. But I knew by reading these 12,000 diaries that managers and leaders sure didn’t act like they knew this progress principle. Their behaviors suggested that they weren’t even really paying attention to the progress their people were able to make day by day. So I tried to think about what can we do to make it clear that this is something people need to pay attention to. And that’s where we came up with the idea of doing this additional, very small study to survey almost 700 business leaders at different levels and give them that simple survey question. Alright: Here are five employee motivators. Rank order them in importance. And what we found was that progress came in dead last in the survey of nearly 700 managers. Only 5% of the respondents rated progress number one. So it’s like they were actively downgrading progress. It’s like they were actively saying, okay, this is not something that matters. I ask them to guess what is the number one employee motivator. And I say, I’ll keep you on a cliffhanger for a little while – I’ll tell you about it in a few minutes. But I will tell you that none of the guesses that the people in this room have just shouted out loud are the answer. Then when people see what I show them, that yes, recognition and reward do matter, the reaction isn’t skeptical.
You list a range of things artificial integrity needs: environmental sustainability, long-term solutions, incorporating ethical considerations, fostering innovation. These are noble aims. But at the moment, the US in particular doesn’t seem to be backing equity, sustainability – the key things that European values share. Does artificial integrity have any real chance of taking hold when the world’s most powerful AI player appears to be moving in the opposite direction?
That’s a very important consideration. The current US posture does indeed emphasize acceleration, infrastructure building and deregulation, and a kind of geopolitical leadership around AI. The White House AI action plan is explicitly framed around winning the race. So the tension you’re talking about is quite real. But that does not mean artificial integrity has no chance. It just means it will not emerge primarily through one government’s idealism. It will emerge through the combined pressure of multiple layers: regional regulation, technical standards, private and nonprofit innovation, sector-specific governance and institutional risk management. The US, even though extremely influential, is not the whole operating system of AI governance worldwide. And actually, the most powerful force is often not Washington – it’s the market.
If you want to sell into a regulated or high-trust market, you follow the rules of that market. GDPR is a good example. The European AI Act effectively exports requirements through market access. So global providers often need to align because it’s cheaper than running two separate governance stacks. So the real contest is not between integrity-led AI and non-integrity-led AI. It is between different models: some more market-accelerative, some more extractive, some more responsible, but all trying to combine innovation with guardrails that fit what society will demand. And those who succeed will be those finding the right fit with that societal demand. Artificial integrity is and will be the condition for durable innovation and the license to operate over time.
Looking at OpenAI replacing Anthropic at the Pentagon – companies that are less encumbered by ethical guidelines seem to be moving the fastest. Why should anyone slow down to embrace ethics and do the right thing?
Because moving fastest and building the most durable advantage are not the same things. Firms willing to accept broader operational latitude may win specific contracts faster, while firms that keep stronger constraints may look slower. But that is only the short game. The deeper question for any company is not how do I remove friction this quarter. It is how do I remain trustable, deployable, governable at scale over time.
In every high-impact technology, actors can gain speed by externalizing risk, but eventually the bill arrives – through failures, backlash, liability, procurement restrictions, reputational damage, political intervention. Anthropic is actually a useful example here. Its responsible scaling policy is a voluntary framework for managing unacceptable risk – what I was calling the purple line they have defined. That may create friction in some deals. But it also signals a governance maturity that more and more public and private actors will increasingly look for. OpenAI may have won the Pentagon contract. But Anthropic won visibility and a wave of public trust – and that is concrete. The Claude app rose to the top of the App Store right after the Pentagon story, while ChatGPT faced a backlash with a lot of uninstalls.
In every high-impact technology, actors can gain speed by externalizing risk, but eventually the bill arrives – through failures, backlash, liability, procurement restrictions, reputational damage, political intervention. Anthropic is actually a useful example here. Its responsible scaling policy is a voluntary framework for managing unacceptable risk – what I was calling the purple line they have defined. That may create friction in some deals. But it also signals a governance maturity that more and more public and private actors will increasingly look for. OpenAI may have won the Pentagon contract. But Anthropic won visibility and a wave of public trust – and that is concrete. The Claude app rose to the top of the App Store right after the Pentagon story, while ChatGPT faced a backlash with a lot of uninstalls.
I was interested to read that and remind myself of the idea that specific skills and mindsets could be more valuable as we move forward. But are new-collar jobs enhancing the human input and role, or are they risking making a segregated or a two-tier workforce?
The idea of new-collar jobs is powerful because it challenges the assumption that capability must always be certified through traditional academic pathways. New-collar jobs can widen access and recognize practical intelligence, learning agility and applied skills alongside formal credentials. But the model has a real ambivalence. If organizations use it seriously, it can democratize opportunity. If they use it opportunistically, it can create a second-tier workforce, people trained narrowly for immediate technical utility but with less mobility, less recognition, less long-term bargaining power. The real question is not whether new-collar jobs are good or bad in themselves. It is whether they expand human capability, or merely repackage labor stratification in modern language.
The idea of new-collar jobs is powerful because it challenges the assumption that capability must always be certified through traditional academic pathways. New-collar jobs can widen access and recognize practical intelligence, learning agility and applied skills alongside formal credentials. But the model has a real ambivalence. If organizations use it seriously, it can democratize opportunity. If they use it opportunistically, it can create a second-tier workforce, people trained narrowly for immediate technical utility but with less mobility, less recognition, less long-term bargaining power. The real question is not whether new-collar jobs are good or bad in themselves. It is whether they expand human capability, or merely repackage labor stratification in modern language.
How do you persuade a business leader that ethical AI is worth it?
Artificial integrity becomes essential when AI becomes economically visible. Because the magic of these systems is not just about their intrinsic performance, it is also about the outcomes generated in interactions with people. The question is whether you prefer to pay upfront through AI design and governance, or pay later through failure, knowing that anything you fail to address upstream will cost more downstream.
If AI starts to shape our decisions, will we find that we’re collaborating with it or slowly deferring to it?
Both are possible. If AI shapes decisions without weakening our ability to question, override and remain accountable, that’s collaboration, and that’s fine. But if AI shapes decisions by making us passive in the process, that’s where the issues come. Artificial integrity is very much about designing systems with the imperative of keeping us on the first path.
“Many actors still treat integrity as a secondary layer added after deployment, rather than a condition by design from the start.”
Is there a risk that human skills atrophy as a result?
Yes, absolutely. There is a risk of skill atrophy and cognitive, social and moral atrophy as well. If we outsource too much of our judgment, that is what is at stake. AI can remove friction, but friction is sometimes how humans learn. Unintended consequences include de-skilling, dependency loops, homogenization of thinking, false confidence, authority drift, degradation of institutional knowledge and fewer human interactions.
How is artificial integrity doing right now, and what can people do individually to embrace it?
Artificial integrity is emerging. We are still in an early phase, and structurally, acceleration remains stronger than integrity. The less comfortable truth is that it is still uneven, reactive and often subordinate to competitive pressure. Many actors still treat integrity as a secondary layer added after deployment, rather than a condition by design from the start. There is that recent move regarding OpenAI, for example. After being publicly scrutinized around teen safety, they announced the rollout of age prediction to identify likely under-18 users and automatically apply extra safeguards. It illustrates how integrity is still often deployed reactively after harms and reputational risk become visible. So that for me illustrates where artificial integrity is today.
Artificial integrity is emerging. We are still in an early phase, and structurally, acceleration remains stronger than integrity. The less comfortable truth is that it is still uneven, reactive and often subordinate to competitive pressure. Many actors still treat integrity as a secondary layer added after deployment, rather than a condition by design from the start. There is that recent move regarding OpenAI, for example. After being publicly scrutinized around teen safety, they announced the rollout of age prediction to identify likely under-18 users and automatically apply extra safeguards. It illustrates how integrity is still often deployed reactively after harms and reputational risk become visible. So that for me illustrates where artificial integrity is today.
-
anchor decisions in layered values:Balance societal, organizational and individual needs through clear guardrails that engage with real-world contexts. -
Responsibility is shared:Progress requires both global coordination and individual accountability. Systems only work when people actively uphold them.
-
Keep things human:Let AI enhance human judgment, not replace it. Agency, critical thinking and interaction are key to avoiding overreliance.
Stay ahead of the curve!
Sign up for our newsletter and let Think:Act bring you up to speed with what is happening today and guide you on what’s happening next.
.svg)