In April 2026, Gadi Evron — founder of AI security firm Knostic and one of the most credible voices working at the intersection of AI and institutional risk — sat down with Anderson Cooper on CNN. Speaking about the pace of AI capability development, Evron relayed and built on an argument being made by Craig Mundie, former Chief Research and Strategy Officer at Microsoft, in senior policy circles. We aren't, Evron told Cooper, birthing just a new tool.
"We're birthing a new species. It's not carbon-based like we are, but silicon-based. It is a new species that we're going to have to learn to control and collaborate with — before it makes us its pet."Gadi Evron, Founder & CEO, Knostic · CNN with Anderson Cooper · "Terrifying Warning Sign," April 8, 2026 · approx. 2:53–3:12
That is not a cybersecurity statement. It is a statement about the nature of human responsibility at a threshold moment. And it sets up a question this essay takes seriously: if AI represents something genuinely new — not a tool, but a collaborator; not a system, but a presence — then what does it mean that the humans responsible for governing it don't yet know enough about it?
This gap isn't based in speculation — it's in the actual data. And it's only getting wider.
The Research Questions
Two Questions, One Underlying Truth
This essay is organized around two research questions that emerged from analyzing the Stanford HAI AI Index 2026. They are designed to be specific enough to answer with data, and consequential enough to matter beyond the data.
Research Questions
Both questions are asking the same underlying thing: is the human infrastructure keeping pace with the AI infrastructure? Evron's framing — a new species, not a new tool — raises the stakes of that question considerably. Because if he's right, the answer matters not just for organizations, but for how human judgment itself evolves over the next decade.
The Setup
AI Is Already in the Building
The numbers from the Stanford HAI AI Index 2026 — published annually by Stanford University's Human-Centered AI Institute, which tracks AI's real-world impact across research, economy, policy, and public opinion — are not predictions. They are descriptions of what is already happening.
62% of 18-to-29-year-olds are using AI in 2025, up from 33% in 2022. Nearly double in three years. 88% of organizations have adopted AI in some form. 58% of the global workforce uses it at work. And AI incidents — documented failures, harms, or near-misses — have tripled since 2022, from 104 annually to 362.
The generation entering leadership pipelines right now have for the most part never made consequential decisions without AI in the room. Think about the last time you had to figure something out on your own — a question about a reading, an assignment you weren't sure how to approach, a work problem you had to think through. Did you reach for AI first? Did you take what it gave you? Did you check whether it was right?
Think of AI like a very capable new colleague who has read everything ever written, never sleeps, and has no ego about being asked basic questions. The issue isn't whether they're in the building. They are. The question is: who's responsible for what they do there?
Finding One — Research Question 1
The Problem Isn't Technical. It's Human.
The HAI Index draws on a McKinsey global survey of organizational leaders to identify what's blocking responsible AI implementation. The results are striking not because of what they show, but because of what they don't show.
The number one barrier isn't budget. It isn't regulatory uncertainty. It isn't the technology itself. It's knowledge and training gaps — cited by 59% of leaders in 2025, up from 51% in 2024. That 8-percentage-point jump is the largest year-over-year increase of any barrier on the list.
The fastest-growing obstacle to responsible AI governance is that the people responsible for governing AI don't know enough about it. And while the knowledge gap is widening, the capacity to respond when things go wrong is shrinking.
In 2024, 28% of organizations rated their AI incident response as "excellent." In 2025, that fell to 18% — a 10-point drop in a single year. The share of organizations experiencing three to five AI incidents per year jumped from 30% to 50%.
More incidents. Worse response. And the self-reported reason, from the leaders themselves, is that they don't have the knowledge.
Imagine a driver whose car gets significantly faster and more complex every year — but whose training has never kept pace, and whose ability to handle emergencies is getting worse, not better. At some point, the gap between what the car can do and what the driver understands stops being a background concern and becomes the most important variable on the road.
That's where we are with AI governance.Finding Two — Research Question 2
The Public Feels the Risk More Clearly Than the Experts Do
Chapter 9 of the HAI Index draws on Pew Research, KPMG, and YouGov surveys across 28 countries to compare how the general public and AI experts perceive AI's impact on specific human cognitive capacities. The question: what percentage of each group worries that AI will negatively affect a given human mental faculty?
On decision-making and problem-solving: 48% of US adults are worried. Only 30% of AI experts share that concern — an 18-point gap. On metacognition — the ability to think about your own thinking: 53% of the public is worried. Only 36% of experts.
This divergence is easy to misread. It's tempting to conclude that the public is less informed and therefore more worried — that expert optimism is the more calibrated view. But that gets it backwards.
The public isn't less informed about AI. They're closer to what it feels like to use it. They're the ones noticing when they reach for AI before they've attempted to think something through themselves. The experts study AI from a distance. The public lives inside it. That proximity isn't a distortion. It's a signal.
A lawyer who participated in Anthropic's 2026 study of 81,000 AI users put it in terms that echo Evron's framing directly: "I use AI to review contracts, save time... and at the same time I fear: am I losing my ability to read by myself? Thinking was the last frontier."
That tension — adoption accelerating, understanding lagging — was the most consistently reported internal conflict among AI users globally. Not optimists versus pessimists. People simultaneously holding both. Which is perhaps exactly what you'd expect from anyone beginning to sense they are learning to coexist with something that, as Evron put it, is not quite a tool anymore.
What This Adds Up To
Four Things the Data Is Telling Us
The knowledge gap is the bottleneck. Not budget. Not the technology itself. 59% of organizational leaders identify their own knowledge and training gaps as the primary barrier to responsible AI — and that number is growing faster than any other obstacle. When the humans in charge are the constraint, that's where the work has to happen. This directly answers Research Question 1: the barriers are human, they are growing, and the most critical one is epistemic.
Governance quality is moving in the wrong direction. AI incidents are rising. The capacity to respond to them is falling. More exposure, less readiness. The gap between capability and governance is not closing on its own.
The public's intuition about cognitive risk is sharper than the expert consensus. On decision-making, metacognition, and deep thinking, ordinary people are consistently more worried than specialists. This directly answers Research Question 2: the gap is real, it is measurable, and it suggests that the people closest to the lived experience of AI adoption are registering something the field has not fully accounted for.
Tomorrow's leaders are forming their AI habits right now. The youngest cohort of workers has seen its AI adoption nearly double in three years. The cognitive dependencies forming now will shape how hard decisions get made for a generation — which makes the knowledge gap not just an organizational problem but a longitudinal one.
The Question the Data Leaves Us With
None of this is an argument against AI. The same HAI Index documents extraordinary things AI is enabling — in medicine, in education, in access to expertise that was previously out of reach for most of the world. The opportunity is real.
But opportunity and readiness are different things. And both research questions point to the same structural gap: organizations are adopting AI faster than their leaders can govern it, and the public — who live inside the adoption — already sense what the experts are slower to measure.
If Evron and Mundie are right — even partially — the stakes of that gap are higher than a training budget problem. If AI is becoming something we collaborate with rather than simply operate, then the question of whether leaders know enough stops being organizational and becomes something more fundamental: do we understand the nature of the relationship we are entering?
What does responsible leadership look like when the primary barrier to governing AI is the leaders' own knowledge gap — and that gap is widening every year? And what happens to human judgment when the people most worried about its erosion are the ones living closest to it?
Those are the questions this data raises. They don't have easy answers. But the first step is acknowledging that the gap is real, it is measured, and it is getting bigger.