wavePod

Episode 353 Deep Dive: River Nygryn | Trust, Test, Transform: Executive Playbook for AI Leadership - KBKAST | Wave AI Podcast Notes

Back to KBKAST

Episode 353 Deep Dive: River Nygryn | Trust, Test, Transform: Executive Playbook for AI Leadership

KBKAST

Wed Feb 04 2026

In this episode, we sit down with River Nygryn, CISO and AI thought leader, as she explores the critical concepts outlined in the executive playbook for AI leadership: Trust, Test, and Transform. River provides a comprehensive overview of AI’s evolution—from its historical roots in early automated machines and neural networks to the development of large language models (LLMs) and generative assistants. She emphasizes the importance of “trust but verify” in deploying AI, warning against overreliance and the risk of diminishing critical thinking skills. River introduces the 4Ds—dull, dangerous, difficult, and dirty work—where AI delivers the greatest value, and cautions about the loss of creativity and authenticity with widespread use of AI-generated content. She encourages organizations to leverage their unique data sets, underscoring that human judgment and oversight are essential for harnessing AI’s transformative opportunities. River is a visionary cybersecurity and technology le...

Summary

KBKAST Episode 353 Deep Dive: River Nygryn | Trust, Test, Transform: Executive Playbook for AI Leadership

Date: February 4, 2026
Host: KBI.Media
Guest: River Nygryn – CISO and AI Thought Leader

Overview

In this episode, River Nygryn joins KBI.Media to explore the strategic realities of artificial intelligence in enterprise environments. The discussion focuses on what AI is (and isn't), the risks and cognitive impact of over-reliance, practical frameworks for responsible use (like the “4 Ds”), and the critical need for ongoing verification and human oversight in AI-driven decision-making. Through practical stories, real-world analogies, and cautionary tales, the two cut through AI hype to offer actionable leadership advice for executives navigating the AI transformation.

Key Discussion Points & Insights

1. Demystifying AI and Its Evolution

[01:26 – 06:18]

Defining AI, Machine Learning, and Deep Learning:
- AI is a broad term for machines performing tasks requiring human intelligence. The field’s roots trace back over a century.
- Machine learning is a subset, focused on algorithms that learn from data.
- Deep learning, a further subset, powers most recent advances.
- LLMs (large language models) are specific deep-learning models designed to process natural language.
Quote:
"In simple terms, AIs or LLMs are mathematical models that emulate cognitive function... They're shiny new tools in conversation. But you always need a human in the loop."
— River [05:14]
AI’s Limitations:
- Cannot feel emotion, make ethical judgments, or perform expert-level tasks without supervision.
- LLMs can “hallucinate” and must always be verified.
Real-world Risk:
- Example cited of a Deloitte government report misattributed by AI-generated content leading to reputational hazards.

2. Cognitive Impact & Over-Reliance on AI

[06:19 – 16:13]

Historical Analogies:
- Calculators, telephones, and GPS all shifted cognitive burdens—but AI has potential for greater cognitive offloading.
- Risk: Losing critical thinking skills and reliance on AI as an unquestioned authority.
Quote:
"The big one—you don’t want to offload your cognitive abilities to AI too much. It should be there to help you."
— River [09:37]
Educational & Social Implications:
- Critical thinking and source verification aren’t innate for digital-first generations and require explicit education.
- Over-reliance can breed laziness, especially as AI-generated answers are increasingly “good enough” and convincing.
Quote:
"It comes across as being so advanced... that can play tricks on your mind. You start to build rapport with an AI system, which is dangerous because you're subconsciously giving a lot of trust."
— River [12:50]
Market and Social Trends:
- Reduced workforces in some sectors due to AI.
- Culture of “making millions from your phone” fosters shortcuts and can be amplified by AI.

3. Practical Framework: The 4 Ds of AI Use

[16:13 – 20:25]

The 4 Ds (borrowed from Sol Rashadi):
1. Dull (mundane, repetitive tasks)
2. Dangerous (put machines in harm’s way)
3. Difficult (high cognitive load, complex computations)
4. Dirty (unsanitary or undesirable human work)
Examples:
- Dull: Customer service chatbots
- Dangerous: Bomb-disposal robots
- Difficult: Algorithmic trading, cybersecurity anomaly detection
- Dirty: Sewer inspection robots, crop drones
Quote:
"Humans have been using technology in this way—across the four Ds—for many years. It's not just limited to AI."
— River [16:44]
Guidance for Leaders:
- Start small—match clear use cases to these categories for successful AI adoption.
- Avoid deploying AI simply for the sake of it without understanding the problem and solution fit.

4. Creativity, Authenticity, and the “Sameness” Risk

[20:25 – 25:05]

Authenticity is the New Differentiator:
- Noted trend: AI-generated content is making online voices sound homogenous.
- Authenticity will become more valuable and obvious in a sea of similarity.
Quote:
"If everyone starts to become the same, you're going to have to differentiate yourself and stand out from the crowd."
— River [22:10]
Advice:
- Remain true to your own voice and leverage AI as augmentation, not replacement, of creative thinking.
Market Correction:
- As audiences tire of homogeneity, those who “stay true to themselves” will stand out.

5. Security by Design: Trust, But Verify

[25:29 – 34:02]

Core Principle:
- Always apply “trust, but verify”—a classic security mindset—to AI outputs and the tools themselves.
Quote:
"I'd encourage anyone using AI technologies: yeah, trust it. Trust that it's very advanced... but it still makes mistakes and you really need to verify all the outputs."
— River [00:00, reinforced at 26:07 & 36:55]
Verification Tactics:
- Use multiple sources or models (e.g., comparing ChatGPT with Grok), check against the open web, and prioritize first-hand/practical testing where possible.
Risk of “Unknown Unknowns”:
- Novices may lack the knowledge to detect or question false answers. The responsibility is on both users and trainers to improve digital literacy.
Efficiency vs. Diligence:
- While AI saves time by synthesizing information, it can never fully replace real-world testing, critical evaluation, or cross-verification.
Personal Example:
- River describes using multiple AIs to help draft robust technical requirements for an app in hours—what would have otherwise taken months.

6. The Future of AI: Data, Productivity, and Democratization

[34:02 – 36:55]

Transformer Architectures:
- River briefly explains how they work in LLMs, improving AI’s ability to understand context and relationships in text.
Quote:
"The valuable part for the future of AI is actually your data. If you want to take advantage of your unique business proposition... data governance is really crucial."
— River [34:54]
Opportunity:
- Unique organizational data will be a major differentiator in leveraging AI at scale.
- AI will democratize skills and education, offering new avenues for productivity and creativity.
Final Thought:
- Emphasizes maintaining human agency:
  "You are the teacher. You are the master of AI. Don't let it become yours."
  — River [37:20]

Notable Quotes & Memorable Moments

On LLMs and Human Oversight:
"They're shiny new tools... but what they can't do is guarantee truth. There was a recent example... quoted incorrect facts and quotes which experts came out saying were completely false."
— River [05:24]
On Laziness and Over-Reliance:
"AI definitely risks amplifying [laziness]... trusting in decisions made but informed by an AI output that was not validated or checked."
— River [15:16]
On Authenticity in the Age of LLMs:
"You'll have people gravitate to you... if you don't start to sound like everyone else... and don't give in to being lazy and offloading cognitive abilities to a machine."
— River [23:46]
On Data Governance:
"Focus on your datasets—they're going to be crucial for whether you're going to be successful in the AI movement or fall behind."
— River [36:55]

Timestamps for Key Segments

| Timestamp | Segment/Topic | |------------|-----------------------------------------------------| | 01:26–06:18| AI basics, evolution, LLMs, limitations | | 06:19–10:51| Over-reliance, cognitive skill loss, social parallels| | 12:00–16:13| Corporate risk, critical thinking, validating AI | | 16:27–20:25| The 4 Ds framework, practical guidance | | 20:25–25:05| Creativity/Authenticity, voice homogenization | | 25:29–34:02| Trust but verify, security by design, real-life stories| | 34:02–36:55| Future of AI, data governance, democratization | | 36:55 | Closing thoughts |

Conclusion & Executive Takeaways

AI is a powerful accelerator for productivity and creativity, but cannot replace human judgment, ethics, or critical thinking.
Leaders should adopt frameworks like the 4 Ds to target high-value areas for AI deployment and avoid blanket adoption.
Over-reliance on AI is a cognitive risk for individuals and a strategic risk for organizations—instituting “trust but verify” is essential.
Authenticity is more valuable than ever in the AI era—both in creative expression and business identity.
The true competitive edge lies in the unique data organizations possess—data governance and strategy will define long-term winners.
Human agency remains paramount: You are the master of AI. Don’t let it become yours.

For more in-depth insights, visit KBI Media for industry news and expert analysis.

Loading summary...

Transcript

A (0:00)

I'd encourage anyone using AI technologies is to, you know, yeah, trust it. Trust that it's very advanced technology. It's got a lot of smarts in it compared to what we've been exposed to in the past. It's come a long way as an industry in its own right, but it still makes mistakes and you really need to verify all the outputs.

B (0:24)

This is katiecast.

C (0:26)

I'll be completely silent.

A (0:27)

As a primary target for ransomware campaigns, security and testing, and more than six feasibility, risk and compliance, we can actually automatically take that data and use it.

C (0:39)

Joining me now is River Nigren, Sizo and AI thought leader. And today we're discussing Trust, Test and transform the executive playbook for AI leadership. So, river, thanks for joining me and welcome.

A (0:51)

Thank you. Great to be here.

C (0:52)

Okay, so I really want to start perhaps with your view in understanding AI now. And this is a big topic and it's one that people are obviously talking about a lot and it's one that I have spoken about a lot on the podcast. But because there's so many different viewpoints, different opinions, different angles, I still think there's a lot of ground to cover. And I'm really keen to maybe understand for myself, like what it can and can't do. And I asked this because this is important as most people, like the general sort of person, likely have played around with ChatGPT, et cetera. But I'm keen to maybe explore beyond that, perhaps areas that people don't know about AI.

A (1:26)

For example, I think it's important to set the foundation of AI and its kind of definition. So AI is a broad term for machines performing tasks that require human intelligence. It's had a history of over 100 years and just recently it's evolved in the outstanding ways that we know and experience today. So many people define it, or you just use AI as an all encompassing word to capture everything. But that's generalizing AI. It's far more nuanced than that. The concept of artificial intelligence, it's not new. It's been around since the early 1900s. There was a Spanish engineer, Leonardo Quevedo, he created a automated chess playing machine. Once it was set up, it didn't need any human intervention. It made legal chess moves and even alerted to illegal ones made as well. Then in 1943, there was a really interesting paper by Warren McCulloch and Walter Pitts, and they modeled artificial neurons. So they were simulating brain like functions. And the paper actually explored the idea that the brain could be understood as a computational system and that introduced the concept of artificial neural networks, now a key technology in modern AI. And then throughout the 50s, the field expanded into artificial neural networks, experimentation with mimicking human problem solving abilities. And in 1956, a group of researchers and engineers, they coined the term artificial intelligence in a workshop they were having. And that's kind of the official birth date of the term AI. So while the concepts and experimentation with machines thinking and automating tasks, it's been around since the 1900s, the field really gained traction in the 50s and 60s. So it had a bit of a winter in the 70s where not much progress happened. And then it picked up again in the 80s, right up until what we know it as today. So AI is probably the broadest term. It covers everything from chess playing programs to autonomous robots. Then you've got machine learning. So that's a subset of AI. It's a field focused on algorithms. They learn patterns from data, make predictions or decisions without being explicitly programmed. So as examples, spam filter, recommendation system. So what Netflix tells you it thinks you'll like Uber Eats, suggesting what you should order next, that's all machine learning. Then there's deep learning, it's a subset of machine learning. It's got many deep layers and that's probably powered the most recent advancements in AI technology. So they're things like speech recognition, image recognition, all of that type of stuff. You know the technology on your phone when you, you know, type in, you want to find all photos of your cat and it pops up with all of them in there. Or you type in the word passport and that file or photo of your passport comes up on your phone. That's thanks to deep learning and LLMs, they're the specific type of machine learning model. So they're based on deep learning and what's called transformer architectures. So they're designed to process human input and then generate human like outputs. So they're trained on huge data sets. They're probably what we know of and what the public knows of and is calling AI. But really they're generative AI assistants or LLM based assistance if you want to be more precise. So the field is really huge is what I'm trying to get at. So there's even things like small language models which are cut down versions of LLMs that are more efficient and they're designed for faster processing or specific tasks. But in simple terms, AIs or LLMs, they're mathematical models that emulate cognitive function. You may have been heard naysayers call LLMs, Google on steroids or advanced predictive text technology. You know, whilst that's accurate, it's pretty overly simplistic explanation of what they are. They're shiny new tool in conversation. But you always need a human in the loop. So they do really well at augmenting the human user to be more productive, explore new ideas, harness new skills that you know, would ordinarily take decades or a university degree to obtain, predicting our tastes and preferences and ultimately they can make the little things a little easier for us. But what they can't do, they can't feel, they can't make ethical judgments in complex scenarios, they can't perform expert level tasks, they can't guarantee truth. They're always making things up. There was a recent example in the Australian Financial Review of a Deloitte misstep for a government organization in Australia. Piece of work was costing $400,000 or more. And the article actually quoted incorrect facts and quotes from educational studies which some of the lecturers and experts came out saying that it was completely false. So it just really strengthens the fact that when it comes to interacting with artificial intelligence, we always need a human in the loop.

A (7:45)

Yeah, definitely. I mean, so I remember those, those maps, those big books, and then, you know, having to turn the page and map out the number from one page and then, you know, the next map was 20 pages into the book. So it's really confusing. I mean, the same thing happened with calculators as well. So there were, there was an uproar from teachers with calculators saying, you know, that students would lose their mental arithmetic abilities. Happened with the phone. So the phone that used to have that kind of round dial where you had to kind of really remember the numbers, that all changed as well. So I think the difference within history, though, like this is this is not unusual. This has happened throughout history where we've adapted technology. It's augmented human capability, and therefore we've lost our skills in something. Now to date, those skills that we've been losing, you could probably argue weren't really advanced. They weren't complex cognitive abilities to do things like that, like remember, you know, where the streets are and all that kind of stuff. But in today's age, AI is a different beast. AI can allow you to, you know, program an app in an afternoon. It can do advanced research for you where, you know, the human brain is inherently lazy and, you know, you may not go and check all of your sources and, you know, explore different avenues for, to find information, whether that be in physical books, journals, talking to people, all that kind of stuff, because AI can just do it at our fingertips. I think there's a lot to be learned from the wave of social media and what happened there. When social media expanded throughout the world and gained mass adoption, I don't think anybody was. Well, at least it might be the same problem. People aren't talking about it enough, but nobody was really talking about it in the public about what it could do for your cognitive abilities. So the pattern and the habit of scrolling and constantly getting dopamine hits has an effect on the brain. There's A lot of evidence to that now and there's a few papers actually out there that talk about the real risk of losing our skillset, more importantly losing our ability to critically think. So that's probably the big one. Like you don't want to offload your cognitive abilities to AI too much. It should be there to help you. But then on the same token, the positive light, interacting with AI and using it more frequently and then comparing how you use it to others using it, your own AI interacts with you. So as you become more educated with prompting and understanding how to get the best out of and the best outcomes out of AI and knowing what follow up questions to ask you, your results can be far superior to the person next to you. It's all about just being aware and being self aware of when you are getting into those negative patterns around just letting AI do everything for you. So again, strengthens the argument like human in a loop. Make sure that you're always validating its output and checking the sources.

A (16:27)

There's a TEA leader in AI, she's from America. Her name's Sol Rashadi. I read this on her blog, actually. And then I did some reflecting around technology throughout history and realized humans have been doing this across the four Ds for many years. So what they are are dull or mundane work, dangerous work. So where you'd rather risk A machine than a human life, Difficult work. So really high cognitive load to get an outcome, or dirty and unsanitary work. So ultimately it's a way to determine what would you like AI to solve for and what should it be used for. So it's a bit of a general guide, but given the fact that we've always bucketed technology within these four areas to solve specific problems, it's probably a good guide to start with. So I'll share some examples, a historical one and then a modern one to shape the story. And you know that humans have been using technology in this way, so it's not just limited to AI, but dull or mundane work. So assembly line robots, that's an example from the past, they revolutionized manufacturing to today where we've got customer service chatbots on websites handling repetitive, frequently asked questions, then you've got dangerous work. So rather risking a machine's life rather than risking a human's life, you'd risk a machine's life. So there's nuclear inspection robots, they use them in Chernobyl and then to current day times there's bomb disposal robots. Difficult work. So weather prediction models, cybersecurity, anomaly detection, or algorithmic trading in finance. So they're all examples of where technology and, you know, AI models have helped humans complete difficult work, then dirty and unsanitary work. So you know, sewer inspection robots in the 90s right through to, you know, Yoroomba that you might have at your home now, and even crop spraying drones in agriculture. Like if you're right here, you could probably expand the D theme to distance as well. So we've used technology to solve the distance problem as well. So from telephones to the Internet to drones doing things for us now.

A (25:59)

It's a old school security principle. Always trust but verify can be applied in so many different facets. So even when it comes down to assessing a vendor or a suitable solution, whether that's AI or not, you know, it's always you go down into your exploratory phase, you start to talk to people to understand what it does, what it can and can't do, and then you go into a proof of concept phase to determine whether it's fit for purpose for you. So that's where kind of I'd encourage anyone using AI technologies is to, you know, yeah, trust it. Trust that it's very advanced technology. It's got a lot of smarts in it compared to what we've been exposed to in the past. It's come a long way as an industry in its own right, but it still makes mistakes. And I think it was Google that they have a link into the, into one of their web browsers that they were still saying it's got to make mistakes. They're only released the new edition of this out to a thousand people, but they acknowledge that it makes mistakes. It's. We're still in this phase of exploring LLMs. Even though they're all new and exciting to the average user, those that are the experts in the fields and those that have been working in this field for decades do understand that they are still in their infancy. We do still need to test them. We need to security test them especially. There's, you know, like any new software or system, there's going to be adversaries trying to break them, there's going to be curious researchers trying to break them, and they have and they report those findings. So that's where I kind of really narrow down onto, yeah, trust them, they're advanced technology, but you really need to verify all the outputs. And if even as simple as if you're getting AI to guide you or teach you something, verify it by actually doing it. So if it's giving you these steps to learn something and it's a practical skill, put it into practice somewhere and actually test that what it told you is accurate and worked.

A (30:51)

I don't think so. I mean personally I like, I use it a lot, you Know, I was a little reluctant at first just because I work in technology. I understand the risks, I was understanding the limitations. I've got a few friends that have been in the AI industry for 40 plus years. So speaking to them, I was always hearing from them, you know what, AI is not. So I was a bit reluctant to try it and then saw another wave of people I know really embracing it and embracing the technology and highlighting different use cases of how great it is and how it can help humanity. And so I gave it a go and had a look. And what I would say it just makes you faster. It's not going to. So for me, if I was trying to find something on the Internet just via a Google search, it's. Google's not going to scour it or suggest something else for me to search or suggest. Maybe I want to explore this topic as well in conjunction with what I'm currently exploring. Whereas AI does that. It's kind of got that predictive technology in IT to kind of guess what am I looking for, what do I want to find, what am I researching? So all of that's a massive time saver. Even though you do still have to verify the outputs, it's probably cut down the work by days or weeks, depending on what task you're trying to do. The other thing for me was I've had aspirations to create my own app for many years and it's always been murky for me. I've overseen app development projects but never been too close to the developers or to understand what they're doing. And the programming languages in the app space change so frequently you become obsolete quite quickly. As a developer in that world, I did an experiment with AI. I had a friend of mine, that's his job, he's got an app development company. I spent four hours on an afternoon just fleshing out my requirements. And this was before some of the more recent models where you've gone now, the models are a lot better at vibe coding. So, you know, speak to it in natural language and it will come up with your requirements. What I did, I actually blended, you know, Gemini, Grok, chatgpt, I went to all of them and kind of, I knew how to build out technical requirements just because I've done it before in my corporate history and work experience. So kind of had that model to work with, building out your functional requirements, your business requirements did that, plugged it in. But I had no idea how to then inform an app developer what to do to make something easier. I proposed a project and wrote it all up, sent it to my friend and he was astounded. He actually said, did I write this? I said, well, no. AI helped. And I was curious because I wanted to know, is it bad? Like, can you work off that or not? And it was about two months of me talking to him on a regular basis, him asking me to provide requirements that I couldn't. It was just taking me too long to even get my head around it and learn, like, what does an app developer need? And yeah, AI in four hours allowed me to do it. He said it was better guidance and better instructions than majority of his institutional clients providing.

A (34:26)

with us today, River, AI models themselves are one thing. There's a lot of mathematical algorithms that are using transformer architecture. And transformer architecture just back in really simple terms is it's just the architecture that LLMs have built on in order to look at a sentence as a whole and then assign attribute weights to the words. In transformer architecture, they're called tokens, but there's weighting attributed to that. So if you've got a sentence and you're talking about a cat and you've got the word it in it, you know that the technology can infer that when it is used, it's referring to the cat. So that type of intelligence is there in the models. However, the more valuable part, and I think there's a lot of business, there's a lot of individuals starting to understand this more and more is the valuable part for the future of AI is actually your data. So when I talk about this at security conferences, I really emphasize the fact that a data governance initiative is really crucial. If you want to take advantage of your unique business proposition or value as an organization, you want to leverage AI to do so. Every organization will have unique data. That data might be valuable, it might not be. But understanding that and how you can leverage AI technology to benefit from that or create revenue streams, that's something that's really exciting where I think the AI industry will eventually go in future. The other thing I think as well, it's going to open up the ability for people who may not have been able to access traditional education. I think it's going to be able to give those individuals real skills that they can test and validate and prove their value as an employee. It's going to democratize the education space to an extent. It's going to make research faster as well, people more productive. At the end of the day, I think it's got a lot of opportunity and potential to really augment humanity, to become more productive, more efficient, and even more creative in some sense. I mean, I'm excited for the future of it, but I think it's just really important to be aware of the risks versus the opportunity and be pragmatic about it.