In this edition of Author Talks, McKinsey Global Publishing’s Adam Volk chats with Tiffany Xingyu Wang, president and cofounder of Oasis Consortium, and World Economic Forum Young Global Leader, about The Trust Code: How to Unlock AI’s Promise Without Losing Humanity (Wiley, September 2026). Wang examines why our instincts for trust can misfire when applied to increasingly capable AI systems. She argues that learning to calibrate trust can help individuals and organizations use AI with intention, preserve human judgment, and capture its benefits without sacrificing agency. An edited version of the conversation follows.
Why did you write this book?
I wrote this book because I kept watching very smart people ask a question that’s almost too big to be useful: “Do we trust AI?” AI’s already here, and we trust it when we use our credit cards, because fraud detection is running behind the scenes. We trust it when we switch on the light, because various types of software are balancing the grid. But more and more often, AI models are drafting our work, and, increasingly, AI agents are acting on our behalf. The central question of our time is no longer whether to trust AI, but how much trust does a system deserve, under which conditions, and for which tasks?
Often, we see that being overly trusting exposes us to harm, and not trusting enough traps us in paralysis when it comes to innovation. It’s only the calibrated trust, which this book advocates for, that enables us to move forward. Professionally, I spent over a decade building trust and safety systems for big enterprises.
Often in boardrooms, crisis rooms, and on conference stages, people ask me what trust means in the AI era. I never really had the time to give a holistic answer. When I began writing this book, I was pregnant with my second child, and my son was two years old. That’s when the stakes became even more personal to me.
My son talked to Alexa and rode in the Waymo with me regularly. None of this is remarkable because the next generation will not remember a time when AI was not already embedded in our daily life, making every decision with us and on our behalf.
We may not control how fast AI is advancing, but one thing we still have the time and the power to control is how deliberately we give AI that authority. I searched for a book that could teach me that missing skill, but I couldn’t find it. So I decided to write it.
Was there anything that surprised you in the research, writing, or response?
We have a machine—cognitive technology—that sounds warm, very confident, very empathetic, but without any of those signals corresponding to understanding, responsibility, or intentionality that a human would have. This creates a profound evolutionary mismatch: the Stone Age brain we have evolved from is evaluating Space Age systems.
That mismatch is one of the first surprises that I encountered during the overall process. It’s very foundational to understanding many aspects the book covers. The second surprise came from writing the book itself. The faster AI moved, the more convinced I became that the book shouldn’t be about what is happening with AI now but about how to think about whatever happens next.
We may not control how fast AI is advancing, but one thing we still have the time and the power to control is how deliberately we give AI that authority.
As you can imagine, it’s terrifying for an author to write about what is probably the fastest-moving technology in history. I wondered whether the book would be outdated at the time it hit bookstores. I tried many ways and spent much effort to really go back to first principles when explaining any concept, any headline in this book, so that people could use it as an evergreen guide to decipher whatever comes next.
Why is it so difficult to define trust when it comes to AI?
After speaking with many people while researching and interviewing, I found the distinction that cleared this up for me.
There is trustworthiness, which is about the system itself. And we can talk about parameters that matter to AI technology. But in this context of AI, is it truthful enough? Is it competent enough? Is it also benevolently aligned with your personal goals or your enterprise goals well enough? Is it appropriate for that context? That’s about the system.
Trust is a different thing. Trust is about us. Am I willing to rely on it?
Another matter is calibration, and whether those two match. A technically very strong system can be under-trusted, and a very charming but unreliable system can be wildly over-trusted. That’s what’s happening everywhere now. Both [approaches] are actually failures.
Instead of asking, “Do I trust AI?,” what we should ask is, “What vulnerability are you accepting today for what promise in the future and in what particular situation?” Because the definition of trust is that you accept vulnerabilities today in hopes of a promise in the future. Trust remains a very vague sentiment until you can articulate and understand the vulnerabilities you accept today and the goals you want to achieve tomorrow for very specific conditions and situations.
But once you correctly articulate the question and think through it—some frameworks we talk about in the book—trust becomes a decision. It becomes a decision that you can examine, measure, and improve upon.
When it comes to AI adoption, which areas of trust do you see leaders neglecting the most?
I really focus on what trust means through the lens of AI. There are four prerequisites for trust: competence, truthfulness, benevolence, and contextual fit. I chose those four because I constantly saw people mistake competence for trustworthiness.
It’s very important to understand that AI is so capable that it will sound very competent and aligned with you, while it has no reason to be benevolent to maximize your well-being. Understanding that pitfall allows you to have the sense of guardrails you need to constantly think about when using AI. Manifestation is the quadrant that is especially important for enterprise leaders now. Does our behavior match what we say we trust? There are so many research papers now being published that say that. We started pilots, but the ROI is low.
The definition of trust is that you accept vulnerabilities today in hopes of a promise in the future.
It’s very important to understand that a pilot is a statement of belief. What really shows a state of trust when an enterprise adopts AI or not is whether we change the workflow, whether we train the staff, or whether we change the incentives. Match what you say to what you do, and think about how you use the signals of behaviors to assess the level of trust you delegate to AI.
Finally, there are the unseen factors. It’s important to understand that these technologies are meant to be warm, empathetic, and very confident. But they are not earning our trust by delivering the track record and the proven evidence, either by ROI or by reliability over time. They’re simply triggering our trust.
How can we hold AI players accountable when even they can’t predict how their systems will behave?
In today’s AI age, accountability means owning the conditions around unpredictability, and that articulation is extremely important. There is a missing actor in this conversation about AI. We usually draw a map of two posts: AI labs that build models, and regulators that write the rules.
But enterprises decide what gets bought and what’s connected to real systems. We give access to real data and deploy it at scale. Consider any mature industry, for example, aviation. Airplane builders are responsible for designing and testing it, but airlines have the responsibility for how the plane is operated and who controls it.
Of course, the regulators set a floor nobody can opt out of. AI needs the same layer. Often, the power and responsibility of enterprises as a whole are underused and underrecognized in AI governance or AI accountability. In other words, enterprise leaders have more agency than we give credit for in this race of AI.
If I’m buying AI for an enterprise, I would stop asking, “Is this model safe?” Instead, I would say, “Show me a failure and show me how you contained it. Show me how quickly I will know [about a failure], and show me what recourse I have if it costs me money or reputation.”
Enterprise leaders need to understand the strong agency you have to decide what model you buy, how you deploy it, and what protection measures you take. Hopefully, when enterprises come together and make that the default, the norm, we can see the maturity of this industry.
Labs build capabilities and regulators set the floor, but it is the enterprise that decides how much authority we give it.
Often, the power and responsibility of enterprises as a whole are underused and underrecognized in AI governance or AI accountability.
Can you give an example of how AI is expanding human judgment and perception?
I interviewed Professor Stephen Quake at Stanford, a pioneer of single-cell genomics. He described biology to me as historically 90 percent experimental and 10 percent computational. The ambition is to reverse it so that it’s 90 percent computational and 10 percent experimental.
We have already seen a glimpse of this with AlphaFold, which used AI to predict more than 20 million protein structures and to understand better how biology works. A cell takes it to a more complex, different level, because every cell has 20,000 protein-coding genes.
Understanding cells better means we can better understand our body, our health, our diseases, and our therapies. But when it involves running in 20,000 dimensions, that’s beyond human brain capacity. That’s something you can’t do without AI.
But that doesn’t mean human judgment is stripped from the process. Instead, where the judgment is situated shifts. Arguably, the more AI can surface and test hypotheses for us—more cheaply, quickly, and at scale—the more important human judgment becomes. Because ultimately, humans must decide which hypotheses to pursue, which tasks and experiments to undertake, and how to apply results in ways that benefit humanity—and weed out harmful applications, such as using that same technology for bioweapons.
Biology is a very clear example of how AI expands our perception that, as humans, we operate in three dimensions. When a task involves high-dimensional data, we need AI’s help. But it makes our human judgment more important.
If you move from science to art, it becomes controversial. In the creative world, many designers and creatives face this question: With AI on the rise, what is the value of creatives? To answer that question, I interviewed two masterminds of creativity and design: Brian Collins and Greg Hahn, both of whom run their own design advertising shops. They are at the top of their game, and no doubt both are using AI. The workflow is becoming like an hourglass. On the top, it’s faster and cheaper to run variations and do research, and at the bottom, it is cheaper and faster to distribute it.
But the narrow waist of the hourglass is our taste, our care for the audiences, and our judgment of what would work, and our understanding of cultures. These are the many things that make one design somehow click with you, and one advertising becomes viral, versus others simply that don’t.
Human judgment continues to be a scarce and valuable asset that we need to practice and own.
How has your personal relationship with AI changed after writing this book?
My personal relationship with AI is becoming intimate but unsentimental. I use AI every day by default, and I trust almost none of it. Writing this book didn’t make me more or less cautious; it simply made me more specific. I started the process expecting to land somewhere on the optimist-to-pessimist spectrum, and I ended up rejecting the spectrum entirely.
I think binary trust is lazy trust. The question is, trust it to do what, under which conditions, and with what recourse? I don’t really have a position on AI anymore. What’s most important now is to have a practice, because trust is not practiced. It’s not well-defined, but the most important skill is being able to protect yourself and your enterprise, and unlock AI’s promise.
As a parent, that’s where trust matters most to me. My children talk to Alexa without hesitation. I’m not trying to keep AI away from them. That ship sailed before they were born.
Judgment isn’t taught to children in a lecture format. Children watch someone practice it, and then they learn it. I’m trying to make sure my children see me make decisions and hear me say, “Let’s check it out.” I’m genuinely optimistic, but not because I think the systems will get safe on their own.
We talked about accountability, the systematic thinking we need to have, and what agency enterprises and individuals need to take on. The variable that determines how this goes is not only the intelligence and capability of machines; it is also whether we, as humans, keep exercising that judgment in their presence.
Unfortunately, I see this less discussed and practiced, and that is the purpose of this book. I trust AI precisely, but not generally. I want my children to inherit the habit of deciding for themselves. That’s really the trust code.
