Carlos Garcia (Head Cybersecurity at Solaris)

We talk with Carlos Garcia, Head of Cyber at Solaris Bank, about leading security through rapid transformation while keeping risk grounded in what can break today. We dig into how AI changes software supply chain security, identity governance, and fraud, and what practical controls still work in a regulated bank.
• moving from cloud-native speed to regulated banking process and why compliance becomes a security foundation 
• using company transformation to build a proactive security posture without losing focus on present-day attacks 
• why AI-assisted coding blurs internal vs third-party code and raises supply chain risk 
• practical supply chain controls: pause auto updates, review new libraries, use internal registries 
• transitive dependencies explained and why SBOM visibility matters for modern AppSec 
• non-human identities, tokens, and why AI agents create new rotation and tracking problems 
• governance and observability for AI agents using existing security signals across code and network 
• assigning human responsibility, audit trails, and “know your agent” as a future trust model 
• evaluating new security vendors for regulated environments beyond the product demo 
• biggest near-term worries: supply chain compromise and AI-driven fraud like voice cloning

Transcript

Welcome And Guest Introduction

Speaker 0:02

Welcome to the Unicorn CISO podcast. This is Pedro from 33N. Let's get started with today's episode. All right, so thanks everyone for joining the new episode of the Unicorn CISO podcast. Today we're very lucky to have uh Carlos Garcia, head of security or head of cyber at Solaris Bank. And I'm sure we'll have a great discussion. Carlos, welcome to the podcast and thanks for the time.

Speaker 1 0:28

Hi Pedro, thank you very much. Thanks for having me on the podcast along with the rest of the fantastic people.

Speaker 0:36

All right, all right. Adding one more for sure now with our conversation.

Career Lessons From Cloud Banking

Speaker 0:41

So I I wanted to start with the the with the question on your career to date and how that shaped the way you think about the cybersecurity leadership role. You spent a long period also at Santander Open Bank before Solaris. What how was that experience? Could you give us an overview on how that translates into how you think today at Solaris?

Speaker 1 1:01

Yeah, yeah, sure. So Aspience was was super interesting because back in 2018, Open Bank uh went along in AWS. So at that time we were essentially operating as a pure vintage. This was like the speedboat of uh Santander Bank, which is a massive bank. In those days, we enjoyed a lot of flexibility. So we could move facts, experiment, uh, build in the cloud. So that period was super super nice, but then we started hitting milestones and began with the international expansion. And this changed the things a little bit because we start facing with uh highly regulation, uh like getting you know approvals, especially in the US. So this leads us to embrace the compliance. And when all of this happens, as a security leader, you learn uh how to appreciate that. Because if I go to to the present now at Solaris, we can I can see some similarities between those early days uh at Open Bank when we operate more on a you know on a flexible way. But yeah, when you start thinking on security, cyber security, realize uh the value of these uh processes that the big organizations have because it gives you a very good foundation. So yeah. If I if I go to Solaris now, that's an ecosystem that is undergoing into a massive uh transformation, and on top of that, the company is making a very aggressive bet on the um AI, trying to put on all the process. So that is a very, very interesting moment to be here.

Speaker 3:11

Indeed, I wanted to touch on on that. And on the business side, there's a huge opportunity for Solaris, and this is this is public information. And and it also, I'm sure, puts puts some some strains or has some consequences to in to the security side. So how

Using Transformation To Reset Security

Speaker 3:27

does this new chapter of transformation affect your priorities and and how different is this from your previous experience?

Speaker 1 3:36

Yeah, I mean, that the transformation is always a challenge when you when you speak about cybersecurity, probably for the rest of the areas, but my top priority right now is to write this wave of internal transformation. So I I see this uh not as something bad, uh just the opposite, because I think that uh if the company is changing, we can also change the way that we put the cyber on on top of that. So I want to use this momentum to build a proactive security posture from the ground. Yeah, so the idea is to build uh a roadmap, but not uh, you know, when you put uh deadlines on a on a roadmap, it's easy to get uh hyper focused on the future, but the cyber attacks happen in the present. So every single second is an opportunity for a break. That's why I think that you need to balance between the long-term cultural transformation, but also to keep keep the company secure in the present. So I'm happy with the moment, but of course, there is a big challenge right now.

Speaker 4:57

Just to give context to our listeners, so Solaris is the embedded finance or the backbone for uh operating banks or for digital banks, the most leaning forward uh ones. And so the whole wave of identifying operations in a bank, right, is where Solaris finding the opportunity, right? And so this is a wave that every business is going through, and and and it's just a big opportunity for someone that is a digital backbone of a bank, right? So so this is this is the way you describe it, right?

Speaker 1 5:26

Yeah, I mean, probably all the comp not probably all the companies are trying to adopt artificial intelligence uh at some way. The challenge is you know to adopt on a highly secure way, especially if if you are a bank, because you are handling money or you are also handling the trust of uh your customers, your partners. So this needs to be done with uh highly security standards.

Speaker 5:51

Yes.

Supply Chain Security In Practice

Speaker 5:52

One of the the the the topics that is also I mean, and in the headlines day in and day out, and it is uh also very relevant for for someone that carries the the um the embedded finance platform of of digital banks is the topic of software supply chain security, right? And and again, in the case of of Solaris being the the the embedded finance platform, it's it's not only that it affects you, but it directly affects your customers, right? Whenever there's there's a bridge, the blast radius extends very directly. How do you see this this the opportunity to contain risk in practice? How do you size this opportunity? You also are coming with a strong background on AppSec before. So how how do you see the context of Solaris?

Speaker 1 6:41

That's a good question, and that's something that I think should keep all the cybersecurity teams uh awake uh at night right now because it's not a new problem. But today I think that is a bigger problem because with the artificial intelligence adoption the line between what is the code is written internally and what codes come from a third party is not clear. It's you know quite blurry because we are seeing this all the days, major organizations being compromised because of a single dependency, and probably that company never knows that that dependency exists on his code. Because if you are using artificial intelligence a lot uh for for make your code, you have that, you know, the the model suggests you okay, we can do this with this third party library, the developers say, okay, why not? Let's uh let's do it. But you are not assessing the security of this third party. So my approach for that will work hand in hand with the developers because if we start stopping things on the pipelines, we are going to have some problems because of that fight. So we have the reputation on cybersecurity teams to be a stopper. I think that we need to collaborate to collaborate on a joint strategy and make the developers part of the security solution. Because this will need a lot of change, cultural change. The first one that uh that my first suggestion for that uh for protect that will be post automatic uh updates. Because we are seeing that maybe you have a very well-known third-party component in your system, but at some point this component was compromised, so you are compromised, and this is not because the component is not big enough, it's not well maintained. Sometimes it was an insider on the supply chain, some other times was a transitive dependency that we can speak about the transitive ones in uh in a moment. But the idea is okay, stay, let's say, a couple of versions behind. So update this this dependencies on a manual way after a review. And connecting with that, you need to do that review with a forest principle. Always, if you want to introduce a new third-party component, just review, be sure that uh the company is trusted, then you can start working with that. And if you introduce uh or if you use a third-party library, try to load from uh internal registries. Instead of use a public registry, this kind of you know NPM websites. It's better to have your internal registry. I know that is uh more hard to maintain because you need to maintain and it's not maintained by the community, but on the other hand, you have the security that uh the specific package that is inside of your repository is a trusted one. So if someone compromise the third-party repo, you are not going to be compromised because it's internal one. So that will be my three advice here in terms of the software supply chain security. Try to keep the control because uh it's the only way that when something appears, you are able to understand if you are affected or not.

Speaker 10:39

I think this is also very relevant because everyone I speak to doesn't see a solution on the quick patching side, right? So the prevention is probably the the fairest solution as much as it can be done, right?

Speaker 1 10:51

So yeah, I mean, especially for the patching side, we have the problem when with the um transitive dependencies, because uh normally, okay, we are using that specific uh package, which is super secure, is well maintained. But what is behind this package? So there is another dependencies, how how this package is built. And this is the real problem, because even the tools that normally uh we have on the market are not always able to see that transitive dependence, or maybe they are able to see that dependencies, but uh until certain level, I mean you cannot see the full the full chain. If uh if I need to put an example of this when I try to explain what is uh transitive, it's like imagine you you are manufacturing a car and you know where your engine are created, where your braking system comes, but then you don't know uh about the teeny the tiny screws, for example. So if those screws are used to connect the the engine and the braking system, you have a problem because if the screw company has a problem, then all of your chain was broken. So this is the same. Imagine that you have a library which is secure, but this library has another 50 libraries behind, and then you have 50 potential problems. So the solution here is to build um S-BOM, a software build of materials, and understand everything behind your code. I know that is more easy to say that to do.

Speaker 12:38

Exactly, especially maintain and and have uh tough relationships.

Securing AI Agents And Machine Tokens

Speaker 12:43

I wanted to touch on on the also the um the non-human identities and identity side because you were also in uh in a in a panel or a presentation uh a few a few months ago. And and uh and and it's it's especially with with that with transformation that uh solar is is leading with identifying a lot of the operations. I mean, this this uh topic also comes, right? Securing agents, the non-human identities, uh management. How does this look in practice? What are how are you thinking about the this this space?

Speaker 1 13:15

Yeah, yeah. For me, that topic is also is not a new topic. I mean, we are being non-human identities long time ago because it sounds like something super super new or super modern, but think on a non-human identity, thinking a token. I mean, we have uh that tokens everywhere. And that's something that when I was speaking on the AIC at Berlin, I make that question and I ask about how many people have token in the system that is not able or they don't want to take the risk to remove or to rotate because maybe break something on production. So that problem is not new, it's long, long time ago. But uh the problem now is more big because if we are start spinning up identities with artificial intelligence, and we are allowing that AI agents on the scene, the things become more complicated because right now, let's say on the traditional workload, we have a process, we have a token, and that token is going to or that process is going to use the token to do, let's say, linear actions. For example, we use that token to do 100 actions and we program to do okay, then do one, then do the next one, like uh let's say first in, first out. But then we start using AI. So we need to remember that the AI is non-deterministic. So we never know how the AI is going to act. For example, that same process that we are mentioned, maybe the AI, if if the agent have the necessary permissions, the agent decides that instead of do one and then another, and then they decide to spin up hundreds of sub-agents with hundreds of permissions, and then do all the action at the same time because it's more convenient for for that. So at that point, you have an agent that is creating sub-agents, so it's creating sub-identities. So, how do you rotate that hundred of new identities created? Or how do you even understand that there is a lot of identities behind? That's the real problem. And also, the other day on the on the conference, I saw JubiKey, you know, these uh MFA tokens like uh the RSA RSA tokens. And I asked, okay, how do you attach this to an agent? You can't, because uh you can attach, you know, Pedro has these tokens, so this MFA, in order to Pedro use his credentials, is a it needs to touch the the button here or it needs to put uh six code, but with an agent, you can't you can do that because it's meant to work uh on an automated way. So I think that the real problems come with all of these uncontrolled actions that you are not able to track. The solution for me, it's clear, it's just have a very good overview about what do you have in terms of agents and especially what capabilities, what permissions that that agents have. Because if you have only read you only have a read-only agent, then okay, you are more secure than the agent is not going to create a super agent, for example. But if the real problem here is that uh all the companies right now want to run and wants to be the first on the AI landscape, so sometimes the people that is creating that uh AI is not uh thinking about the securities, just thinking about having the process the first. So that's the real problem for me.

Speaker 17:31

And and in the context of of Solaris, you see that uh the vendors you you've been seeing and and and both emerging vendors as well as your big A Bless House, right? So I'm sure there's some solutions you also explored with the larger vendors. Where do you see the maturity of the market here, right? Can you find the the solutions you need for this governance and observability of age?

Speaker 1 17:57

Yeah, yeah, I think uh there is a lot of um you know a lot of solutions right now for that. The interesting part here is that uh the solutions work very different, each one, which is not bad, it's just a different approach. So I wouldn't say that uh we have central solution, probably that that was the best thing to have, a central one, but almost all the solutions that we already have in place for other things are incorporated in this AI governance part. So, for example, if you have a solution to do static analysis on the code, they are able to see the code and see if you are making API calls to third-party AI solutions like uh open AI or cloud or whatever. So then you you have a source, then you you know that your developers are trying to connect to LLM. Then if you have a tool that is meant to control the network, you are able to detect that there are network requests to that websites. So I would say that is not yet mature in terms of uh okay governance for AI, but the market is moving on that direction for sure.

Speaker 19:24

Very, very clear.

Liability And “Know Your Agent”

Speaker 19:26

One one of uh the uh still within the the topic of of identification, one of the areas of interest is also whenever you're talking about authorization, auditability, liability, so you're mentioning MFA that this attaches a human, and so if something goes wrong, this this person is liable, right? In the context of uh platform that is as core as a banking uh core platform, uh how how what is the thinking on around liability and if there's something that goes wrong, how can you identify uh well the the the chain of of approval and trust there?

Speaker 1 20:03

Yeah, that's very interesting topic, and it's a conversation that I have more than once here because I think that there is a gap on that space. Uh I'm going first with the basics. So the basics, if you have an agent inside of your company, then you should have a human responsible for that agent. Because it's the only way that, okay, that agent is doing that. Who is the responsible? And I'm not talking about who is the one that is uh coding the agent or who is the one that is giving the permission. No, no, no. You need to have a compliance document that says, okay, the agent 123 is created for do this action with these permissions, and the final uh responsible is that person. You need to assign that responsibilities. In my opinion, right now is the only way that you can you can do you can do that, or even you can have this, let's say, security gates on the agent. So the agent needs to stop at some certain point at get the human intervention on the loop. And then if you have the human, then the human is the responsible for. But thinking about that, I was thinking more on uh know your agent, not for not for companies or not for internal process, more on a global perspective, because we are a banking company, so all the banks uh we will have the know your customer. So is this something that is not new, and all the banks need to do this assessed uh first prior to to create a new a new client. But what happened when the customer is starting? Not being a human is being an agent. So for me, that part will be very interesting. And I think that this is a new area where we can explore. I think like something like you know, Sufa here in Germany or the Survey in Spain, we have a database of uh drafted agents. If we are able to create something like this, I mean I'm not saying that uh we are, but the one that is going to create that, or if a company sees that opportunity, probably there is a you know good space here for create something because right now there is people, but on let's say one year there will be agents acting behalf in behalf of some people. So the companies we need to understand who is behind because uh if you want to open an account, but it's not Pedro, it's Pedro's agent. Who is the responsible? Is Pedro the final responsible? Will be the agent, will be the IT third party. So probably we need to have this central governance part and not inside of the company, let's say inside of a country or yeah, probably a country or inside the European Union, something that is able to give us a risk score or a you know quality score of the agent.

Speaker 23:33

I'm sure there will have to be a solution, and then that we had the insurance for cyber attacks, right, in the past, right? So I'm sure there'll be uh even non-cyber industries benefiting from this. So that's

Buying Security Tools In Regulated Banks

Speaker 23:46

interesting. I wanted to touch also on on uh other emerging topics that you might be seeing and and beyond this ones that we covered on the software supply chain security or or non-human identities and the and the genetic security. What are things that keep you interested that you think are even if not hot and in in the news that you you believe in that delivered around for you and that that you see value?

Speaker 1 24:12

I mean it's is it's it's hard uh today to filter the noise because and and this is a good question because uh today a couple of developers can spend some money on some AI like cloud code and create something that uh looks super nice, and probably that uh is it's a really good idea, and probably they can get some money and start selling. But what I found normally with this case is that these small companies start failing when need to go into a highly regulated environment. Because it's not only about the product, it's only about all the service that surrounds the product. Uh, because right here in a bank, we need uh SLAs, we need the company that we have an enterprise grade, the security operation center, incident response, pen tests. So when you put all these new companies, the hyperloop hyper hyper tools through a CPRM process, then they are not able to pass. So for me, when I start looking into new new tooling and new companies, I need to look a little bit more far, not just about the idea, but also about the company. So something that uh probably no one do is I try to look who is putting the money behind. So, for example, I think that this is a very good role of a VC like a you, like because probably you are able to bring that knowledge to the new companies in terms of okay, the product is super good, the idea is super good, but then you need to create uh things surrounding this. So I wouldn't say that I I want to speak about any specific product because there is a lot, especially in the cyber area. But uh yeah, I would say that uh that will be a suggestion for for that new new products that is not only focused on the product, it's all focused also on the enterprise level of the of the product.

Speaker 26:41

When we speak to founders, there's always many of them that uh look into the financial services as kind of one of the biggest opportunities. And and I think people uh may overlook sometimes the regulatory side. I'm sure you've you've also gone through it, at least I've always been surprised with how much of uh an effort the compliance and regulatory side requires, especially in financial services. So it's definitely um in agreement

Biggest Near Term Risks And Fraud

Speaker 27:05

there. I wanted to end with uh a question on what what are your concerns, what worries you more in a few in in the next few years? And it's it's all of them are quite concerning, but uh if you think about the level of fraud, if you think about the identity abuse from agents, if you think about software supply chain security, what are some of the concerns you have on the immediate term, right?

Speaker 1 27:29

Yeah, you never know because um this probably on this world you start looking on something that uh you think that this is your biggest problem, and then something super small cause the big problem for you. But uh, I would say that probably the software supply chain uh compromise because, as we are saying, it's hard to track what you are putting inside, and especially now with the artificial intelligence, the development happens super fast. All the people want to create something, even the people that is not technical people, which is fantastic, but also this brings you problem because right now the the marketing guy sees a couple of videos on YouTube and then open cloud and say, Okay, I create a platform, so we don't need to pay anymore for that SAP system that we have because I create my own, which looks super nice, but again, on behalf of what is supporting that? So the software super chain, I think that is a biggest concern, but also the fraud, because I'm not super concerned about you know this mythos model, a glass wheel project. I think that is quite uh noise here, but I'm concerned about the new capabilities that the bad actors have in terms of the fraud area, because right now they are able to create very good clones of the original website, but also they can create the fakes, they can combine the old techniques like uh you know the um uh call spoofing with this uh voice cloning, so the attacks will start becoming super super good created. So the fraud bar is uh is uh also another topic that I think that will be hot topic with the AI on the next uh on the next days. Uh actually it is it is right now.

Closing Thoughts

Speaker 29:38

Amazing. Carlos, really appreciate uh the insights. This was uh very practical, very uh insightful. Thanks once again for joining the podcast.