Menu
Sign In Search Podcasts Charts People & Topics Add Podcast API Blog Pricing
Podcast Image

CISO Series Podcast

AI Confidence: It's a Trap! (LIVE in San Francisco)

05 May 2026

Transcription

Chapter 1: What mistakes have been made in security management?

0.031 - 13.189 Sara Madden

Biggest mistake I ever made in security. Go. Saying I told you so. There's nothing wrong with mistakes as long as it doesn't have a material impact on the business. I love the saying, there's no losses, there's only lessons.

0

13.849 - 26.669 Sara Madden

But when I think back across my career and I think about the mistakes that haunt me at night, it's the times that I have been to the pressure of business and I went against my better judgment. And I put myself in a position of you can see around corners, you know, it's going to happen.

0

27.11 - 34.966 Sara Madden

And when that does and you say, I told you so, it's never a good feeling because it's not our jobs to say, I told you so. It's our jobs to manage risk and never have that happen.

0

36.208 - 43.302 Unknown

It's time to begin the CISO Series podcast recorded in front of a live audience in San Francisco.

0

52.867 - 76.353 David Spark

All right. Welcome, everybody. Welcome to the CISO Series podcast. I am David Spark. I am the producer of the CISO Series. And look, this guy to my left, you may know him. It is the CISO for Rivian and my co-host since day one, Mike Johnson. Let's hear it for him. Hi, everyone. What an amazing audience. We have a huge crowd here today. This is great. This is great.

76.373 - 99.026 David Spark

Now, by the way, we are at B-Sides San Francisco 2026. Let's hear it for B-Sides. And this is our fourth, I believe our fourth time doing a live show here. I think that's right, yeah. Not for you personally, because I know you missed one. This is my third. Number three. We've done four live shows here. Thrilled to be back. This is pretty spectacular.

99.046 - 120.251 David Spark

For those of you listening, we are in a gorgeous movie theater with our giant logo and all our sponsors logo, which I do want to mention right now. Our three sponsors, and I want to hear a huge round of applause for them. Our sponsors, Nudge Security, Quiller AI, and Zenity. Let's hear it for all three of them. We are going to hear a lot more about them in just a minute.

120.331 - 134.33 David Spark

Now, I want to bring our guest in, who is our far left. We had her on once before. So thrilled with how well she did before I said, you've got to come join us live for our show at B-Sides. Big round of applause for the CISO of Convera, Sarah Madden. Let's hear it for Sarah.

135.352 - 136.333 Sara Madden

Hey, everybody.

Chapter 2: How can vendor selection impact security decisions?

198.419 - 216.58 David Spark

And what, do you drop the CISO series? Do you drop Rivian? What is it you drop? Well, I drop that I know David. And does that not get you kicked out? It usually does, yeah. It does get you kicked. It's never been successful. By the way, it doesn't count if they say it because they've had a concussion. Do you know who I am? Or ask, you know, what year is it and who's president?

0

216.62 - 218.663 David Spark

Because I don't know who I am, yes.

0

218.683 - 219.804 Sara Madden

You're a very important person.

0

220.726 - 222.268 David Spark

There you go. I believe I am.

0

225.311 - 227.114 Unknown

Is this the best use of my money?

230.518 - 250.341 David Spark

Quote, when you only see 5% of the options, you can't make the best choice. You can only make the best choice from what you saw. End quote. Here's a pattern, noted Richard Stennin of IT Harvest, evaluate three or four vendors, pick one with the best demo, sign a three-year contract.

250.861 - 267.367 David Spark

Now, 18 months later, rip it out because the vendor got acquired and product development stopped or a better solution emerged or the market leader turned out to be the marketing leader. Now, the implementation time is wasted. And don't forget the squandered political capital spent justifying the original decision.

267.968 - 289.157 David Spark

Due diligence isn't just about evaluating vendors, but knowing which vendors to evaluate in the first place. So, Mike, I'm going to start with you. What does your vendor selection process look like to avoid this trap? And second, how do you balance the fear, if you have it, of making the wrong choice against the need to make a decision and just move forward?

289.938 - 294.704 Mike Johnson

Well, the first step is don't sign a three-year deal the first time that you're working with a vendor.

Chapter 3: What strategies can be used to maximize conference attendance?

438.967 - 452.409 Sara Madden

hire and deploy all the controls. And you're not going to mess with that in the middle of a one or two year license. And so came up for renewal last year with most of our three year vendors, and I put them in either one or two year licenses intentionally for that that reason.

0

453.07 - 472.443 Sara Madden

But you run into a situation where you have to continually air your grievances with your vendors to force them to do the right things. And you leverage the stick of I won't renew, right? We're in challenging times where the majority of the tool sets that we have are SaaS products and they're going up and down. They're unavailable a lot.

0

472.463 - 487.89 Sara Madden

I think product development has taken a shift where we used to do N minus one because you don't trust the latest release. And we've got tons of vendors now that are saying, don't trust until I have like a preferred release. And that could be months, if not quarters, if not nine months out.

0

487.87 - 508.467 Sara Madden

So I've been pressing a lot of vendors lately of just sticking to fundamentals of software development and being confident about the software that you ship to us. And if you get into a situation with your vendors where they're shipping product that isn't stable, isn't good, is causing issues, I'm all over them all the time now because I think we're taking a shift into... low quality releases.

0

508.547 - 524.852 Sara Madden

And this is across most of the vendors we have in our tool set right now. It's happening across the board. So I'm pushing on that a lot as a CISO because I don't want that to be like a new normal. We can't get ourselves into a situation where we used to be an N minus one and we thought that was okay. And then we're dragging further beyond that. Like that's not a good position for us to be in.

525.272 - 543.257 Sara Madden

Right. The other thing I'll say in terms of finding good tool sets is it's not, it's also the peer, I agree with everything that you said, but some of the best tool sets we have, I've learned about them from new engineers that are entry level people that got to test out cool tools in their prior job and they're bringing in really cool new tool sets. And it's as much about learning.

543.237 - 555.896 Sara Madden

finding the right tool as it is making sure that you have the skill sets on your team to manage that. Oftentimes we pick tools because the engineers we have on our team have expertise in being able to manage that product because you're never going to get a perfect product, right? There's always going to be issues.

556.277 - 564.229 Sara Madden

But if you don't have the talent to manage it, you could have the greatest tool set and you could buy and it's not going to work because you don't have the right engineering to manage it. So it's about people too. It's about their skill sets.

567.033 - 570.038 Unknown

Is AI going to help us or hurt us?

Chapter 4: How does AI create a false sense of confidence?

650.733 - 670.11 Sara Madden

So we use AI every day, especially in SecOps. Like it's so efficient for us to run a particular IDE tool in a very big cloud, fork some code, throw it in there, have it analyze it, we get results back in five minutes, that's contextualized, and it's actually super helpful. And then that usually starts like a little bit of a panic and a spin cycle. And we like, oh, my God, look at these results.

0

670.15 - 684.703 Sara Madden

And then we don't trust it. Initially, we go and we look at it. And you end up finding that there's pieces of it that's super helpful. And there's areas that you could tweak. And it's usually not as bad as what the tool says, because it likes to be flashy and dramatic, right? So you just don't trust it. And you look at the results. And to me, it's not

0

684.683 - 695.079 Sara Madden

too much different than our regular vulnerability testing that we do with the various different tools we have or the pen testers that we use. There's false positives in almost everything we work on. So just simply don't trust it.

0

695.539 - 704.633 Sara Madden

And then when it comes to just use cases that we're building out, the human in the loop control is just super important until we get more confident with the accuracy of the AI output.

0

704.613 - 707.998 David Spark

All right. Mike, same philosophy with your team?

708.879 - 732.059 Mike Johnson

And I think that really is the new normal. Six months ago, it was, hey, these things are really confident. I'm just going to blindly believe them. But nowadays, everybody understands that you need to double check. And that is the evolution that we've seen. This airs in May. I imagine things have even changed by then. And so we will continue to have that moving forward.

732.841 - 757.43 Mike Johnson

And we will learn along the way. And Sarah mentioned false positives is something that we actually work with a lot in security. That is not new for us. That is normal. And we should make sure that we're reminding our teams like, yeah, you need to check these things work. And you're seeing this in the industry where AWS had an outage recently that was AI induced, however you want to call it.

758.152 - 769.99 Mike Johnson

And they're making sure that they have senior engineers checking AI generated code going forward. And these are things that we're going to continue to learn. And yeah, nobody trusts these, but that's okay.

772.568 - 777.876 Unknown

Who's our sponsor this week?

Chapter 5: What are the challenges of trusting AI outputs?

1009.409 - 1016.078 Mike Johnson

Well, the first scenario is something is about to happen. The second one is something has.

0

1016.698 - 1020.503 David Spark

Has happened or something is very visibly weak.

0

1020.844 - 1034.421 Mike Johnson

But both of them are something will happen. It's a matter of time. It's a matter of time. And then one is the issue. And frankly, I think the problem that you've built a team around two people, that actually is the worst scenario.

0

1034.401 - 1036.929 David Spark

Well, no, no, it isn't just two. There's a bunch, but if two go.

0

1036.949 - 1040.118 Mike Johnson

But you've essentially built it such that. Built it that if two go, it's going to be.

1040.138 - 1045.153 David Spark

So I think first scenario is the worst. First is the worst. All right, we're throwing this to you, Sarah. What do you think?

1045.252 - 1050.462 Sara Madden

I agree with you. The reason you've got unmanaged service accounts is because you only have three people.

1051.424 - 1060.401 David Spark

So if you lose any of those, you're probably... Well, it's a difference. In the second one, it may not be that you have exhausted heroes in the second scenario. It's just a completely different scenario.

1060.441 - 1077.627 Sara Madden

Well, if you lose your resources, then you can't fix the technical problems, which is the principle that I agree with. Right. I think the worst scenario is losing your good people when you already have a small team, for sure. I mean, none of us have been in perfect environments where we don't have the second scenario. Let's just be honest with ourselves.

Chapter 6: How can organizations effectively manage AI risks?

1246.372 - 1247.634 David Spark

We're going to play another one. Here we go.

0

1248.415 - 1250.638 Unknown

Taking care of personnel issues.

0

1251.66 - 1265.681 David Spark

Okay. There's a hint of being correct here. There's definitely a hint of being correct. Oh, wait. Let them answer first. Hold on. I don't know. Human resources? No. It's a cyber term.

0

1267.043 - 1268.665 Mike Johnson

Security awareness. No.

0

1269.406 - 1271.329 David Spark

I give up. Take a stab.

1271.579 - 1272.68 Sara Madden

This is Thump the CISO.

1272.701 - 1276.526 Unknown

This is not what is Dave's mom. Taking care of personnel issues.

1277.267 - 1278.348 David Spark

And just a hint.

1278.368 - 1279.209 Unknown

Account termination?

Chapter 7: What role do tabletops play in incident response?

1634.314 - 1635.936 Mike Johnson

So it could be all of them.

0

1635.956 - 1638.98 David Spark

I would think and or is what you'd say. Go on. And or. I'm just giving you crap.

0

1639 - 1642.084 Mike Johnson

Go ahead. Thank you. Thank you. Never happened before. No.

0

1642.104 - 1642.704 David Spark

This is the first.

0

1643.265 - 1665.841 Mike Johnson

Mark date and time. Yes. But it really does come down to there's an investment the company is making in that person. That's great. They're advancing, but we'd like to see them bring that back to the rest of the team. So have a debrief, have a document that you write up. Here's what I saw. Here's what I learned. Here are the talks that I sat through that you should go watch the recording of.

1666.963 - 1674.557 Mike Johnson

Bring that back to the rest of the team, along with the plan going in of what you want to accomplish. And that's how you get the most out of these events.

1675.195 - 1677.079 David Spark

Sarah, what do you do with your team?

1677.099 - 1696.26 Sara Madden

I mean, I send my team to the technical conferences so they can sit and learn and then come back and do a brown bag. So like to your point, it's like purpose built, right? Besides Black Hat, AWS conference, like those are good technical conferences that our teams go to learn stuff and then come back and share what they learned. We frankly send the people on the team that like to do conferences.

1696.3 - 1710.18 Sara Madden

There's people that don't like to do conferences, right? And so you can incentivize them with different kinds of training. And so you send the people that actually are going to get out there, go to the talks, meet people, learn things, and then come back and share it. So I think that's another important part of it, too, is like send the people that are going to get the most out of it.

Chapter 8: How can companies ensure vendor accountability during crises?

2229.993 - 2235.567 David Spark

This was from Rishay Joshi of Elementrix who gave this question. Anything to add to this, Sarah?

0

2235.918 - 2251.852 Sara Madden

I mean, we're thinking about how to measure effectiveness of AI and how much our engineers are using AI. And we're looking at what metrics make sense. And it can't be like a conflict of interest in a kind of scenario. So you have to look at like the goals for AI that you have as a business and what those outputs are and then measure it that way.

0

2252.053 - 2267.516 Sara Madden

If it's engagement with the particular ID tool that you're using in engineering and you want to make sure people are using tokens, sure. But obviously it has to just be proportionate to what you want to do for your business. But I mean, it's important like... There are a lot of engineers that don't want to adopt AI because they think they're going to displace themselves.

0

2267.576 - 2274.379 Sara Madden

So what are the kinds of AI use cases do you have going on in your environment that you can incentivize engineers to do and then roll with that?

0

2274.781 - 2300.485 David Spark

By the way, I hate to break it to you. I just realized all these questions have to do with AI. So get ready. Shocking. I know, shocking. All right. So this comes from Colin Dupre of Run Zero. So agentic AI was introduced last year in a significant way, significant way. It's still the big story this year. What do you believe has changed significantly with agentic AI from last year to this year?

2301.325 - 2306.01 Sara Madden

Adoption, speed, accuracy, reinforcement learning models are getting better.

2306.851 - 2329.476 Mike Johnson

What I'd say is you've seen more the concept of agents controlling other agents. It used to be this is an agent that is standalone. It can do many things. I think what we've seen is smaller but more agents along with some sort of master control agent on top of that. And that's what I think we'll continue to see more of.

2330.435 - 2352.055 David Spark

All right. This comes from Jordan Comma Morado. Again, another AI question. And, you know, there was a little bit of scare in the market with Claude's announcement with security. So Jordan's question is, do you see yourself phasing vendors out because of AI? You're smiling, Sarah. What?

2352.423 - 2363.631 Sara Madden

I mean, every time we renew a software license, we ask ourselves, can we replace this with AI? And I already made that call in a vendor a couple weeks ago. So it's evolving fast. So yes, for sure.

Comments

There are no comments yet.

Please log in to write the first comment.