Noel Russell (13:00)

Foreign.

Dave Buettner (13:10)

And now a word from our sponsor, Threat Locker. Keeping your system secure shouldn't mean constantly reacting to threats. Threat Locker helps you take a different approach by giving you full control over what software can run in your environment. If it's not approved, it doesn't run. Simple as that. It's a way to stop ransomware and other attacks before they start, without adding extra complexity to your day. See how ThreatLocker can help you lock down your environment at www.threatlocker.com. let's be real. Navigating security compliance can feel like assembling IKEA furniture without the instructions. You know you need it, but it takes forever and you're never quite sure if you've done it right. That's where Vanta comes in. Vanta is a trust management platform that automates up to 90% of the work for frameworks like SoC2, ISO 27001 and HIPAA. Getting you audit ready in weeks, not months. Whether you're a founder, an engineer, or managing IT and security for the first time, Vanta helps you prove your security posture without taking over your Life. More than 10,000 companies, including names like Atlassian and Quora, trust Vanta to monitor compliance, streamline risk, and speed up security reviews by up to five times and the roi, A recent IDC report found Vanta saves businesses over half a million dollars a year and pays for itself in just three months. For a limited time, you can get $1,000 off vanta@vanta.com cyber that's v a n t a dot com cyber on today's threat Vector segment, host David Moulton speaks with Noel Russell, CEO of the AI Leadership Institute about how to scale responsible AI in the enterprise.

David Moulton (15:26)

Hi, I'm David Moulton, host of the Threat Vector podcast where we discuss pressing cybersecurity threats and resilience and uncover insights into the latest industry trends. In my latest episode, I sat down with Noel Russell, Founder and Chief AI Officer at AI Leadership Institute, to talk about how to scale responsible AI in the enterprise. Noel's advice? Be a doer, not a talker. In a world racing to adopt AI, It's a reminder that hands on experience matters more than hype and that early decisions about accuracy, fairness and security can have long lasting consequences. This episode will help you ask better questions, close blind spots and move forward with confidence. Check out the episode wherever you listen to podcasts.

Noel Russell (16:12)

Foreign.

David Moulton (16:18)

Noel is a multi award winning futurist and an executive AI strategist whose career spans roles at Amazon, Alexa, aws, Microsoft, IBM, Accenture and npr. And now she's the author of a powerful new book, Scaling Responsible AI From Enthusiasm to Execution, where she outlines the framework and principles that organizations can use to scale AI ethically, securely and effectively. I downloaded the PDF copy of the book and got into it as far as I could before I said, you know what, I need to have a conversation with you about it. And today we're going to talk about AI leadership going from prototyping into production and then how organizations can rapidly adopt what they're doing in generative AI. And what is the tipping point that balancing innovation with risk, speed and responsibility. So Noel, your book Scaling Responsible AI from Enthusiasm to Execution. I think it's already making waves and I especially liked your baby tiger metaphor. And I see you've got your baby tiger with you today.

Noel Russell (17:31)

Bruiser.

David Moulton (17:31)

Bruiser. I love the framing. It's both cute, but you know baby tigers are dangerous if mishandled. Can you tell Us where that analogy came from and what you want business leaders to take away from that analogy.

Noel Russell (17:45)

Absolutely. It actually came from my journey, as you mentioned. Yes, I've worked at a lot of companies. The interesting thing about my career is that I always at the, I end up at these companies before they've done a thing, before they've gone into the world of Amazon Alexa or before they've launched cognitive services at Microsoft. And so I was at Microsoft, I was hired to help the research organization Productize AI. So they had 17 research models that were going to be in my purview. And I immediately thought of them like I would use the term herding cats. And so herding cats kind of transformed into this concept of a tiger because cats aren't that fierce and I'm a cat owner, but you know, like, you don't want a bunch of cats around, but they're more a nuisance than like a danger.

Dave Buettner (18:26)

Right.

Noel Russell (18:27)

And so I realized, like, I needed to change that a little bit. And so we ended up with a tiger. And that metaphor though, has now become even more interesting over time because now we're looking at. You know, I always will tell people, when you start an AI project, you start with this like adorable, cute little model that you think, you know, it does novel things, trite things, it's exciting, everyone loves it, people want to be on the team. And then at some point you're hoping someone will go, wow, baby tiger. Like how big are you going to be? Or what are you going to eat? Or you have razor sharp teeth. Like, how much do you have to eat? Where are you going to live? What happened when I don't want you anymore? Like, no one asked that in baby tiger mode. And so that's how this book was created was literally I was like, what happens when, like it's still a baby tiger but like nobody's asking these questions.

David Moulton (19:14)

So what happens when it grows up?

Noel Russell (19:15)

Yes. How do we, you know, avoid that? Yeah. Baby tigers become big tigers and big tigers eat people. Right. Like so.

David Moulton (19:22)

Yep.

Noel Russell (19:23)

Let's be careful.

David Moulton (19:26)

Well, let's, let's talk about the, the human element of responsible AI. You emphasize that people, not just the technology, are the key to responsible AI. What's the role of a security culture in helping AI succeed at scale?

Noel Russell (19:42)

So in this case, you know, we kind of look at that weaving. I like that you said the DNA. I haven't used that analogy in a while, but it is like it has to be part of the DNA, has to be woven into the fabric of these projects. So now all of a sudden, which is why most of the time, the technology part is probably 25% of what I do when I go to an organization, help them build a solution or deploy a solution. The tech is usually not the hard, hard part. The hard part is how do you get a team of people that are going to care about all the things that we've shared that are going to care about accuracy and fairness and security, and how do you get them into that project early enough to ensure that you've built it into the model's behavior, not just bundled it on. That's why governance is required, but it's not enough, because you can just change your governance policies or worse, get acquired by a company that completely dismantles your governance process. Then what are you going to do? So it needs to be built in. And that's the beauty of having LLMs as part of your infrastructure. So I'll encourage, you know, you just. If we expand our mind and think about how do we use an LLM to actually be the security auditor in these systems and embed it into the deployed feature. So now when you get that feature, an LLMs built in, to say, oh, no, these are the rules by which I'm abided.

David Moulton (20:54)

Yeah.

Noel Russell (20:55)

And there's a framework called the AI Safety system, and Microsoft and Amazon both use it. I think Microsoft's the only one that's kind of called it out, that this is what they do intentionally. But that safety system is like four layers. And it starts with the human AI experience, which is like, that's when you involve security, legal compliance, everyone's in the room, plus the line of business owners, plus the engineers. And you're like, what are we trying to do? And this is when you define delegation, what's the AI going to do? What are the humans going to do? This is like the Skynet moment, right? When you decide if you want to can you give everything to the AI? You could. It'll hurt you, baby tiger, right? But most organizations are like, no, there's stuff I want to keep. And usually security is one of those things. Accuracy is one, fairness is one. So there are certain things. But once that human AI experience is defined, that's not a technical problem, that's like a designer problem. So you have these user experience designers designing how AI will be integrated into a workflow or process. The next thing is the system prompt is realizing with every machine you deploy, you have the ability to control, control the way it operates. Most people, when they think prompt engineering, they're thinking the prompts they use to ask their questions. But this is the prompt that's used to tell the bot how to answer the questions. And that's completely controlled. And most context windows for that. It's like 375,000 characters. That's a lot of space for you to. And I go in. That's the first thing I do in an executive briefing when they're like, yeah, we're using AI. I'm like, great, let's take a look at one. And I go into the configuration of the system prompt, and it's like, you are a bot that does blah, blah, blah. Does it? And I'm pretty sure it's a default setting. I mean, it's not uncommon to many of these security things. You walk in, you're like, you know, we wrote a book on this. There's a document on this. Like, it's well documented, but people just won't do it. Many reasons. Time, resources. But now you can build an LLM that will infuse it into the life of your systems. And feature releases like, there's no excuses now and then, just quickly. The last two are less controlled. One model selection. So we talked about helm, right? Picking the right model for the right task. And then the last one is infrastructure, which, again, we're getting deeper and deeper. So if you're not building a model, you won't get to choose the infrastructure it runs on. But you should know, like, are you running on Amazon? Are you running on Microsoft? Are you running on Google? Are you running on hardware in your basement? Are you good at that? Have you ever built a NIC card? Like, nobody you know asks these questions.

David Moulton (23:19)

Yeah, how far down the stack do you want to go? But you should know.

Noel Russell (23:22)

But you should know, like, or at least they should be transparent about it. Like, even if they. They have what are called system cards. So I was just speaking with the CISO at Anthropic and Meta at the event here, and they both were like, we have system cards. And they monitor how many people read them. And it's like less than 1% of people who use their systems go to that page and download their system cards. Not because they didn't publish it, not because they didn't say, we're responsible. Here you go. Explainability. People aren't even asking the question foreign.

David Moulton (23:57)

If you like what you've heard, catch the full episode now in your Threat Vector podcast feed. It's called how to scale responsible AI in the enterprise, released May 6.

Dave Buettner (24:17)

And be sure to check out the complete Threat Vector podcast right here on the N2K CyberWire Network or wherever you get your favorite podcasts. What's the common denominator in security incidents? Escalations and lateral movement. When a privileged account is compromised, attackers can seize control of critical assets with bad directory hygiene and years of technical debt. Identity Attack Paths are easy targets for threat actors to exploit, but hard for defenders to detect. This poses risk in active directory, Entra ID and hybrid configurations. Identity leaders are reducing such risks with Attack Path Management. You can learn how Attack Path Management is connecting identity and security teams while reducing risk with Bloodhound Enterprise powered by Spectrops. Head to Spectrops IO today to learn more. Spectrops See your attack paths the way adversaries do and finally, a recent episode of the Search Engine podcast tackled an absurd but very real iOS bug. Say the phrase Dave and Busters in an audio message and poof. The message vanishes into the void. It never reaches the recipient, leaving only a ghostly typing animation behind. It's all thanks to iOS's hypervigilant blast door service. Turns out the transcription engine hears Dave and Busters transcribes it with an ampersand and forgets to properly escape it. In xhtml, the poor parser sees the rogue ampersand, panics, and nopes out crashing the message. Basically, Apple's message security is so strict it breaks over the mention of a popular sports bar. The bug isn't dangerous, it's actually a sign that Blast Door is doing its job. But still, maybe don't invite anyone to Dave and Buster's via voice message unless you want your plans to mysteriously disappear. And that's the Cyber Wire. For links to all of today's stories, check out our daily briefing@thecyberwire.com we'd love to know what you think of this podcast. Your feedback ensures we deliver the insights that keep you a step ahead in the rapidly changing world of cybersecurity. If you like our show, please share a rating and review in your favorite podcast app. Please also fill out the survey in the show notes or send an email to cyberwire2k.com Nduquet's senior producer is Alice Carruth. Our Cyberwire producer is Liz Stokes. We're mixed by Trey Hester with original music and sound design by Elliot Keltzman. Our executive producer is Jennifer Ibin. Peter Kilpe is our publisher, and I'm Dave Bittner. Thanks for listening. We'll see you back here tomorrow. And now, a word from our sponsor. Spy Cloud Identity is the new battleground and attackers are exploiting stolen identities to infiltrate your organization, traditional defenses can't keep up. Spy Cloud's Holistic Identity Threat Protection helps security teams uncover and automatically remediate hidden exposures across your users from breaches, malware and phishing to neutralize identity based threats like account takeover, fraud and ransomware. Don't let invisible threats compromise your business. Get your free corporate Darknet exposure report@spycloud.com cyberwire and see what attackers already know. That's spycloud.com cyberwire.