Vibes, Vectors, and Very Real Threats

Scott: Welcome to Tech Interruption, where we break up your day’s routine.

Austin: Join us as we unpack today’s industry disruptions in just ten minutes.

Scott: It’s the ultimate tech talk challenge!

[ Three, two, one… Here we go. ]

Scott: Hi, I’m Scott Hoge, Vice President of Cybersecurity here at ECS. I support growth and delivery efforts in the cyber domain across the company and have a long history of that on both sides of the fence.

Austin: I’m Doctor Austin Amaya. Here at ECS I am Director of AI and Analytics. I am a mathematician and a recovering academic. I have more than a decade of experience working on AI and ML in the DOW in general, and in the Intel community.

Scott: Spooky stuff.

Austin: Absolutely the spookiest!

[ reveals the spooky tooth ]

Scott: So Austin… One of the things I’ve been thinking a lot recently are the cool things that are going on in cyber, my area.

Austin: So much.

Scott: AI for you. And one of the things I’m really so excited about is this move towards preemptive cyber security. Really, for the first time, we’re getting ahead of the curve. We’re getting ahead of the bad guy. We’re getting to the left of boom…

Austin: Yeah. Where we can go and make predictions on what bad things are going to happen. Go fix those things in the environment before we actually get there. The capabilities we have are not just predict it, but rack and stack. This is the most important thing.

Scott: I got to fix this now in my environment because the bad guys are coming for it.

Austin: They absolutely are. They always are.

Scott: Can’t do it without AI.

Austin: Speaking of AI, that’s one of the things that’s forcing you to get faster. So it’s… it’s kind of this funny thing where AI is on both sides. It is both increasing our capabilities as the good guys, but it’s increasing the capabilities of the bad guys. One of the things that I think is the coolest thing in AI right now, that is a case in point is agentic coding or vibe coding. Where you can just kind of go in using some of these tools we have right now and just kind of explain with a plain English prompt, “I want an app that does this.” And then, you let it go, and it’ll spin up an entire repository for you. It’ll go through, it’ll make it, and then you come back a little bit later…

Scott: Write comments.

Austin: It… absolutely… it’s better at making comments than I have ever been. Full stop. Right. And then you come back and you have an entire app. I think that’s cool as heck. But then the problem is, if I can come in here and say I need this app in English, then the bad guy can come in here and say, I need an app that goes “boom” in Russian. And that’s the problem that you have to face.

Scott: Right! Now, we can also use the same kind of AI tools to go and say, “Hey, this is coming. I need to be able to get ahead of that. I got to predict when that’s going to happen so I can go and fix that.”

Austin: Because in the end, what we are trying to do is we are trying to ensure that we, the good guys, stay ahead of the bad guys.

[ TOPIC 2 ]

Austin: And speaking of staying ahead of the bad guys…(something that I know keeps you up at night) Something that keeps me up at night is just how fast AI is changing.

Scott: Oh my goodness!

Austin: The way that the tools we have are changing. The way that the capabilities are changing. and how that…changes the scale at which we need to think about our problems and how quickly that scale is changing. The magnitude and pace is just…I don’t know how to grapple with that. The best of us are thinking about how to grapple with that. That is the kind of thing that keeps me up at night. How do I make certain that we stay on pace of the scale of AI and everything else?

Scott: Well, and as those systems grow and…grow in scale, we’ve got more things now I’ve got to go and protect.

Austin: Absolutely you do.

Scott: I’ve got more servers. I got the new data centers coming up. All of the things. There’s just this explosion of information, sensors, to go and get into all of our… our systems so that we can go and understand what’s going on, where there’s bad guys doing things in the environment, really looking for that…It’s finding a needle in a haystack of needles.

Austin: They’re all needles.

Scott: They’re ALL needles! But I need a good AI to be able to go and find that. Humans are great at a lot of things, but it’s…it’s finding that little thing that looks so much like everything else going on, at scale.

Austin: Yeah.

Scott: And then having all the controls to go and protect that and grow out where we’re going and do that… safely and securely.

Austin: And say, “This is the one that’s important.” “This is the one to pay attention to.” And that leads into the other thing that I worry about, which is, as you have these tools that you’re using to protect, right, you have to give them the autonomy and the agency to go through and say, “I’m going to go through this.” “I’m going to do that.” “I’m going to poke this needle to see if it’s a needle or not.” As these tools are doing more and more for us, how do we make sure they stay in alignment with what we need? How do we put guardrails on them? How do we assure them, and how do we do this at pace? Oh man! That’s keeping me up at night.

Scott: And keeping us safe.

[ TOPIC 3 ]

Austin: So you know, Scott…

Scott: What?

Austin: What I’ve been doing lately?

Scott: Okay.

Austin: Playing Hades 2.

Scott: Oh, great game.

Austin: And one of the things about Hades 2 got me thinking about narrative art. Because of the way that they construct the story that you go through and sort of how random it is and how you go back and forth. And like, the story is fundamentally not linear. I started thinking…What’s the best way to convey a narrative? Is it things like your social media feed? Is it things like books? Is it things like film, TV series, or is it video games? And I gotta say, I’m kind of leaning towards video games.

Scott: You know, I love a good book. I love a good…getting lost in a good movie and stuff like that. But you get a really good video game, and I’ll sit down for a video game for a couple, 3-4 hours, and just that time is gone! And you can do that over and over and over again over weeks and months, sometimes, on some of these big games!

Austin: Yeah! And sometimes the narrative will kind of linger with you, or sometimes it changes…Some of the rogue-like games are really good at that. You have a very different narrative every time you play it. Like one of my favorites are the Fromsoft games, like Elden Ring. So Elden Ring very famously open game. You’re riding across the field in this enormous world on Torrent, your horse. You know, you’re going through and just kind of like finding out what’s over this next hill. And the narrative that you construct through playing Elden Ring is completely your narrative. It’s about what you are doing and what you find next, and like where you decide to go.

Scott: Yeah.

Austin: And like my experience if I play it is completely different from your experience that the bosses that I choose to fight, the adventures that I choose to have, are completely different than what they’re going to be from you.

Scott: Right!

Austin: Which makes it really different from like a movie where someone set up the story and you go from this to this to this to this in linear order, something like Elden Ring…It’s like one framework that we all have unique stories in.

Scott: Yeah, I know for me I’m more of a Skyrim player. You know, play the magic user.

Austin: Absolutely.

Scott: And get you know, there’s a story set up kind of at the front and you kind of follow that through. You’ve got kind of some guide-rails on it. You still have a lot of things you can go and find, but it’s a little more scripted out than what that is. But I think video games are a really great way to tell a good story.

Austin: All right, so we’re agreed?

Scott: We’re agreed!

Austin: It’s video games?

Scott: I’ll take video.

Austin: It’s video games.

Scott: All right.

[ TOPIC 4 ]

Scott: Austin, one of the things I get asked all the time is, “How do I keep up?”

Austin: All the time.

Scott: There’s so many things going on both in Cyber and AI. How do I go and keep up with what’s going on in the industry? It’s a real big challenge! There is such a breadth of…data and information coming into us. News sources, podcasts… journals,

Austin: Yeah.

Scott: The bleeding edge stuff!

Austin: Techniques, attacks, everything yesterday.

Scott: All of that kind of stuff.

Austin: Everything changes from yesterday to today.

Scott: How do I keep up? So I know for me, what I do…I have some newsletters that I keep up on. What’s going on kind of this week. What are the big trends going on, like preemptive cyber security? Where are we going? What are the large things going on…so we can keep the ship…

Austin: So, following that specific topic.

Scott: going in the right direction? But unless it’s like, hey, there’s a big attack that’s happening right now.

Austin: Yeah, of course! That preempts everything!

Scott: Let’s get that out. What do you find on the AI side?

Austin: Similar in AI…The big thing is like if a new, big model drops or something like that, you have to stay on top of that.

Scott: Right.

Austin: But for me, the way I see it is when you’re trying to keep up with how the industry is changing, if you’re trying to stay on the bleeding edge of the changes for all of the industry…Like, you just can’t do that.

Scott: It’s massive!

Austin: It would be your entire job all the time! So, the way I think of it is there’s a trade off. Like you can try to stay on the bleeding edge…On your X. On your Twitter. Like reading arXiv directly. Or on the other end is it like textbooks? And by the time it’s a book, no one cares anymore.

Scott: It’s out of date!

Austin: Right. So you kind of gotta get in the middle. And so for me, the happy medium is, like aggregated stories, like podcasts. I’m big on podcasts and YouTube video because someone else aggregates the information. I view it at double speed, right? And I kind of go through and it keeps me up to date. I’m trading off because I’m not necessarily, “This happened today.” But I can be like, this happened in the last week. One of my favorites is “Last Week in AI” or this is what happened in the last month and I try to trade off that way. So I keep the overview even though I’m not necessarily on the bleeding edge,
but it helps me understand where the industry is and where it’s going.

[ TOPIC 5 ]

Scott: So, one of the things I think about a lot.

[pondering the imponderable deeper]
[pondering the imponderable deeper and deeper]
[pondering the imponderable deeper and deeper and deeper]
[too deep!!! Pull up!!!!!]

Scott: Now with all this growth of AI, is now I have something else I got to go and protect.

Austin: It’s just one more thing.

Scott: It’s just, you know, one more thing! Not only have just all that infrastructure that it grows on, but I have all the data…I have all that training data that I got to go and protect. I got the model itself, which is really the core part I have to protect that. I got to make sure that the right people are getting access to it, that ONLY the right people are getting access to it, that they can do the right things in the system. It’s just one whole new layer of things that I got to go and worry about….thanks.

Austin: Yeah. You’re welcome!! AI has become so critical, so fast. It’s capabilities have absolutely exploded that it’s become crucial to protect it.

Scott: Right.

Austin: And in order to protect it, you have to understand it. What has to be protected when you say “protecting AI”.

Scott: Right.

Austin: I did a video with our colleague, Matt McDonald where we talked a little bit about what that means, where protecting the AI doesn’t just mean protecting the weights or protecting the servers that are on it, but sometimes you have to even protect access to it, prevent people from being able to ask too many questions of it, because then they can kind of copy it. They can figure out what it did, what it does.

Scott: Well, not only that, I got to make sure that some bad guy doesn’t get in there to go and change what it’s trying to govern and do.

Austin: If you can start… If you have access to it. If you have access to the training data, for example… If I’m the adversary accessing your training data… I can go in here and say, “This tank that you are really worried about… That you’re trying to find. “I’m going to change it in the training data so that I label it as ‘The friendliest, fluffiest Golden Retriever you’ve ever seen!’ “That’s not a threat! “This is not the tank you’re looking for.” So, you have to not just protect the AI, you have to protect the data and the way it’s trained.

Scott: Yeah… So much more to worry about. Just a whole ‘nother layer. Now, it’s great we’ve got things covered. We’re going and… and keeping an eye on those things.

Austin: We understand the Cyber… We understand the AI…

Scott: Don’t forget…

Austin: We understand

Both in Unison: we got to protect it.

[ Outro Plays ]