Snowflake Summit 2026 | Mayank Upadhyay, Snowflake & Jason Merrick, Tenable

Clips
More from Snowflake Summit 2026

Mayank Upadhyay

Chief Security & Trust Officer

Snowflake

Jason Merrick

SVP of Product

Tenable

play_circle_outline Tenable and Snowflake: Building a Centralized Security Data Lake Amid Post‑Acquisition Consolidation Challenges

play_circle_outline Tenable’s exposure management versus traditional vulnerability management

play_circle_outline Mythos and frontier models creating rapid vulnerability discovery (vulnerability tsunami)

play_circle_outline Necessity of AI governance, automation, and data protection foundations

Info

Mayank Upadhyay, Snowflake & Jason Merrick, Tenable

Mayank Upadhyay

Chief Security & Trust Officer Snowflake

Jason Merrick

SVP of Product Tenable

search

Dave Vellante

>> Here we go.

Rebecca Knight

>> Good afternoon everyone and welcome back to theCUBE's Live coverage of the Snowflake Summit here in San Francisco. I'm your host, Rebecca Knight, alongside Dave Vellante, co-founder of theCUBE. We've got a new guest for the show. I'd like to welcome Jason Merrick, Senior Vice President of Product at Tenable. Welcome, Jason.

Jason Merrick

>> Thank you very much for having me.

Rebecca Knight

>> And Mayank Upadhyay, chief security and trust officer at Snowflake. Thanks so much for coming on.

Mayank Upadhyay

>> Thank you, Rebecca. Thank you, Dave. Real pleasure to be here.

Rebecca Knight

>> Yes.

Dave Vellante

>> Good to have you.

Rebecca Knight

>> So Tenable, of course, is a cybersecurity company. Why don't you start by walking through what kinds of data challenges you were facing as an organization before adopting Snowflake and how it has helped you face those challenges?

Jason Merrick

>> Yeah, absolutely. So Tenable focuses on exposure management. So if you think about the digital assets that companies are responsible for, what Tenable has focused on is helping build an inventory. Sounds very simple, but having an understanding of what cloud assets, IT infrastructure, web services, identities. Oh, AI is also part of that attack surface. So how do you build that inventory and then analyzing that inventory, looking for toxic combinations, misconfigurations and vulnerabilities? And one of the challenges we had as we were going through and acquiring, I think we acquired seven or eight companies in nine years. So being able to bring all that information together and exposure management is about bringing context, having an understanding of the asset, the relationship within the environment, and then being able to help prioritize that for an organization. What are the critical few things? So we had a choice. The choice was, do we build our own data lake? Do we build our own data pipeline? Or do we find a partner that is the best in the industry of doing that? And so we partnered with Snowflake as part of the Tenable One basically data infrastructure.

Dave Vellante

>> So it is essentially your security data lake, if I could use that term.

Jason Merrick

>> Yeah. It's where we feed all of our information and that's where we do our analytics, that's where we do our prioritization and help our customers.

Dave Vellante

>> You used the term, Jason, exposure management. Is that different from posture management or is it just good marketing?

Jason Merrick

>> No, actually it is a program. Exposure management is not a product. If you think about where our roots are as Tenable, it was vulnerability management. So being able to scan servers, look for vulnerabilities and then help prioritize those. Exposure management is looking at the bigger picture. It's not just looking at traditional IT infrastructure. It's looking at cloud, which is CNAPP, which has a component of posture management. Exposures are about being able to link together the asset itself, regardless of what it is, could be a cloud, could be an identity, but lacing it together context within the organization, looking at the configuration, looking at it doesn't have known CVEs on it? Does that have multiple CVEs? So being able to go and highlight that and then be able to reduce that risk. And in this world of Mythos and frontier based models, it accelerates it. No longer can humans keep up with machines, you've got to be able to orchestrate that agentically.

Rebecca Knight

>> So Mayank, when you're hearing Jason talk about these things and then talk about the before Snowflake and the decision making that went into it, what does it signal to you about how other enterprises are thinking about these very same questions?

Mayank Upadhyay

>> Yeah. No, I think Snowflake is definitely a highly trusted brand, which was one of the things which attracted me to come and work here, especially taking care of security and trust. And it signals to me that they get a lot of value out of putting their data in us, not just from the security and trust aspect, but also on the capabilities we bring for data processing, for building that context Jason was talking about, so then they can serve their customers safely, securely, and super efficiently.

Dave Vellante

>> How-

Mayank Upadhyay

>> It's even more important in the world of AI now. Yeah.

Dave Vellante

>> Well, I wanted to go there. Talk about how AI has changed security. I mean, it used to be we could stop 99% of the attacks and the humans could deal with the other 1%. AI has changed that, moving at machine speeds. Just discuss that and what does it mean from a ... How do you attack that from a data standpoint?

Mayank Upadhyay

>> So back in the day when you build software, you would have different components in the software. As an engineer, you knew that this component was going to call this one API in the next component. Then it was going to munch that data and then call the next API there. It was super predictable. But now in the agentic era, these agents are wiring their sequence automatically and you have no idea how this chain is going to flow. So all of a sudden there's all this new exposure, new data paths which things are going to take, and it's completely unpredictable. It's non-deterministic. So from a security perspective, you kind of have to take care of your basics and then have a lot of observability and then that's how you control this, because it's impossible for you to be able to sort of predict exactly what's going to happen each time you run it.

Dave Vellante

>> So I'm glad you brought up observability because you remember the days of Hadoop. We probably could have done it all in Oracle, but it was too expensive so we created Hadoop, which is dump all the data in Hadoop and we created data swamps. I know it didn't really work out that well. But nonetheless, you needed a lower cost platform to do that. Now, we have agents creating all this data. There'll be massive amounts of data. I think people underestimate the importance of observability. How are you observing agents? I mean, you acquired Observe.

Mayank Upadhyay

>> Yeah, that's right.

Dave Vellante

>> It seems to be a really important part of data and now security.

Mayank Upadhyay

>> Let me talk about that. So there's this whole thing, sprawl of shadow AI where people are bringing up MCP servers all over the place, like metaphorically under their desk. And then when developers want to use them, they go down, developers are frustrated. CISOs are nervous because they don't know who's talking to what and what data is being exfiltrated. So the first thing you want to do is you want to put things in a common choke point, have what's called an MCP gateway and have these agents route through them. So this is the reason we announced our intention to acquire Natoma, which is a company that supports over 100 different SaaS apps. And so we believe this is a way that our customers are going to be able to create this paved road for all of their developers. So when they want to talk with AI to different SaaS services, super easy for them and their CISOs can sort of have that observability we're talking about.

Dave Vellante

>> Do you have a perspective on that, Jason? I mean-

Jason Merrick

>> It's interesting. We're in unprecedented times. So you've got every single company feels they're behind. So AI adoption is just rampant and every board member saying, "We got to move as fast as possible with AI," and security is an afterthought. So you've got all this adoption that's happening. And then you've also got these new frontier models that are able to go through and find vulnerabilities. I mean, you look at what Mythos has done, it has really changed the game of vulnerability. So the CISO's caught in the middle. What's winding up happening is Mythos was able to detect 271 vulnerabilities in Firefox. The previous model, four or five. So massive number of new vulnerabilities are going to have to deal with. At the end of the day, the customer still has to deal with that. So CISOs are struggling with, well, how do I handle the employee use case where employees can go and create agents? I can go into Copilot. I can create an agent that touches sensitive data. I'm an employee. I'm not a security person. I may not know how to configure it. So you've got that risk around governance, visibility, observability. And then on the vulnerability side, how are you going to deal with this vulnerability tsunami that's going to happen that you're going to have to go and deal with? And it's no longer can patching in seven days work, you've got to do it in real time.

Dave Vellante

>> So when I talk to security pros, because Mythos obviously created a lot of awareness across the industry, governments, et cetera. They say not only does it expose vulnerabilities, but it lays out the tech path, and that's what's scaring people mostly. Now I know it's not just Anthropic with Mythos. Every major frontier model vendor has similar capabilities. I've seen Oracle responded by saying basically upgrade to autonomous. I saw Red Hat and IBM came out with sort of this open source approach that they're using, both of them kind of monetizing. Have you guys changed the way in which you're advising customers to secure platforms, the protocols, patching sequences? Maybe it's different in Snowflake, but what's changed in terms of your advice to customers as a result of these new capabilities?

Mayank Upadhyay

>> Look, definitely the way things used to work before in the industry has to change. You have to be continuously scanning and continuously remediating. You don't have the option to wait on that. There's a stat I saw where the time to exploit from when a CVE was released used to be of the order of days and now they're saying it's down to an hour or even half an hour. So you don't have time, right? So when you find these vulnerabilities, you almost have to have the AI immediately generate a patch and then have somebody quickly review it, have a human in the loop, especially if it's complicated and then get it out the door. So I think the rate at which software is patched is going to increase. It's probably going to be happening almost on a daily, if not a weekly basis. It's no longer going to be biweekly or monthly or whatever. I think things have to go out continuously. There are many companies in the industry which we're already doing that. I think everybody else is going to have to catch up to that.

Dave Vellante

>> So that speed of exfiltration that Mayank just mentioned, it's not your swim lane necessarily, Jason, other than you are part of defense in depth. How do you use data to improve what you do best?

Jason Merrick

>> So we're moving to a model from detection to preemption. We have to be able to help organizations understand the digital assets they're responsible for across the attack surface, then look for vulnerabilities, look for misconfigurations, look for toxic combinations, help with the orchestration and workflow necessary. And then we just launched our agentic framework called Hexa. And Hexa's focus is helping with that AI remediation automation capability, and validation is the critical other component that validate that the remediation has happened, the patch solves the issue.

Rebecca Knight

>> But in terms of helping the organizations understand there are still so many blind spots to that, Jason. This is objectively terrifying what you're describing here.

Jason Merrick

>> Yeah. Well, I think there's a multitude of things that we're helping organizations with, particularly the CISOs. I think first and foremost, it starts with an AI governance board. We've always recommended you've got to have a fair use policy and employee policy to be able to go through and say, "Hey, what can you do?" I think that there's also some awareness that needs to take place. Shadow AI is a real thing. Tenable helps organizations. We have a capability called AI Aware. So we can actually find shadow AI in your environment, whether that's in the cloud, whether it's in a web service, whether it's externally facing, whether it's in your internal infrastructure or your endpoints, be able to give you visibility into that. And then from there, it's about helping with actionability. How do we take that information? How do we help solve that issue, mitigate risk for the organization? And then at the end of the day, the CISO really has three personas. They have to manage up to the board, to executive management, and they have to do that at the right altitude and they have to have confidence in the information. So we help be able to provide that in almost, I don't mean to oversimplify, but a letter grade, a traffic signal. Not just a list of vulnerabilities that they have to be concerned about. But then their second persona is how do they work with their peer groups? How do they go to the engineering team, the operations organization, the IT team? There, they have to deal with accurate information. They have to know, yes, this is the issue, this is why the remediation needs to take place. And in some cases, it may impact business. May take a system down for a little bit. And then the third piece is managing their own teams. Unfortunately, security professionals, we lack the talent. So you've got to be able to do more with unfortunately the team that you have. So how do we automate that? How do we get the practitioner out of the drudgery? And these are the three things that Tenable really focuses on for organizations.

Dave Vellante

>> How is the structure of defense, from an organizational standpoint, changing? And what I mean is in the cloud era, way back in the cloud era, cloud was like the first line of defense, we had the shared responsibility model. But then developers were being asked to secure infrastructure, shift left. And then you obviously had the SecOps team, you had IT, you had audit, it's sort of the last line of defense. The shared responsibility model with agentic is now changing. So how are organizations or how should organizations respond?

Mayank Upadhyay

>> So I'll say two things. So one of the things we did at Snowflake is we brought both the IT side of the house where our traditional security team, the security practitioners are sided, as well as the security engineering side of the house together because I think the boundaries are getting blurred. First of all, the vendor landscape may not evolve as fast as the threats are, so many things have to be built in-house. So that's one of the reasons you kind of want to bring these teams together. Also, I think AI is a great neutralizer where maybe if you were not a software engineer before, now you can use an agent and you can develop exactly what you need. So the first thing is the organization needs to step back and take a look at what their security or structure is like and see how these two organizations can be brought more close together. The second thing I'd say is Jason was talking about not having enough talent where even if you wanted to scale, there's no way, even if you had the people, you wouldn't be able to scale at the speed at which AI might be attacking you. So the way to do that is through automation. You have to use automation and AI to fight AI. So that is the other big thing security teams have to be looking at right now. It's just a no-brainer.

Rebecca Knight

>> Think like the bad guys.

Mayank Upadhyay

>> Exactly, exactly.

Dave Vellante

>> Well, you have these new roles emerging, the forward deploy AI engineer, which is, now add that in, that individual is going to have to be thinking about security. I think you mentioned these, you can't just bolt it on as an afterthought, right?

Mayank Upadhyay

>> Yeah.

Jason Merrick

>> Yeah. And we're seeing a trend that a lot of organizations are bringing the security team together where prior it was the identity team, the cloud security team, the vulnerability management team, the endpoint security team. We're actually seeing organizations bring those together. Exposure management is one of the titles that they're actually wrapping around those organizations. So we're actually seeing large Fortune 500 companies actually building out exposure management program, which kind of handles what I'll call pre-breach and then detection response.

Dave Vellante

>> And I guess as you just described that, those specialized roles, it made sense at the time because they had to go deep into their own domain, but then what you got was you got data stove pipes.

Jason Merrick

>> Yeah.

Dave Vellante

>> I mean, there was these islands of specialization which now have to come together.

Jason Merrick

>> Yeah.

Mayank Upadhyay

>> And we spoke about a lot of these new things, changes have to be made, but a lot of the basics still have to be taken care of because fundamentally this is all built on a foundation of data. So all the hygiene around making sure you're protecting your data, you've got the right controls and governance around it, whether it's data encryption, the right RBACs, you name it. All those things have to be in place because if that foundation is not strong, agents are going to exploit it.

Dave Vellante

>> Okay. So the what doesn't change, but does the how you do it change?

Mayank Upadhyay

>> Yeah. Yeah, absolutely. In fact, you have to use AI to process these things. You have to use AI to maybe ... I'll give you an example. A person who works at a company a very long time just accumulates more and more permissions to do things. Now in their job today, they may not be doing all of that. But the moment they get an agent to do something, the agent will explore all possible paths. So if you don't want any sort of incidental damage to happen from that, you better clean up that person's permissions. Not just that, when the person sets an agent to work on their behalf, you better give them the right construct so the agent has its own delegated identity and it says you're allowed to do exactly this one thing and nothing more than that. An agent is a little bit like an intern. You give them a credit card and say, "Go buy a printer from me," and they come back with a massive refrigerator and you're like, "What just happened?"

Dave Vellante

>> BMW.

Mayank Upadhyay

>> Exactly. Or a BMW. So those are the sorts of things enterprises have to grapple with. And it's really, really important what platforms you choose and the platforms have all the right constructs for you to be able to do these things.

Rebecca Knight

>> So a lot of enterprise security teams though are overwhelmed by risk and they're under-resourced too. So what is your advice on where they should focus their energy first? Let's start with you, Jason.

Jason Merrick

>> I keep coming back to this, you have to have an understanding of the attack surface you're responsible for. So being able to build that inventory and understanding of identities as that great equalizer that Mayank talked about, you've got to make certain you've got a clean hygienic directory that has the right entitlements. But understanding that inventory, understanding the risk of that inventory and then putting together a program because you have to now move at machine speed. You have to combat AI with AI and reduce risk. I think that's the change that's happening in the industry right now.

Mayank Upadhyay

>> I mean, it's a fascinating world right now. If you look at what is AI security all about, you can break it down into three parts. At the core of it, you've got the LLMs, you want to make sure the LLMs don't get attacked. So last year, a lot of people worked on prompt injection, and so there's products around that. There's really cool stuff you can do today. Outside of that, there's things like we spoke about identity, MCP governance, things which are part of the next layer around the LLM, as the LLM interacts with the external world. And outside of that, you've got your data, which is where you've got to make sure you have the right governance and the right foundation from the beginning. So that's really how you break it down. You got to take care of all three layers.

Rebecca Knight

>> Excellent. Well, Mayank and Jason, thank you both so much. A really interesting conversation.

Dave Vellante

>> Thank you for coming on.

Mayank Upadhyay

>> Thanks so much for having us. Yeah.

Jason Merrick

>> Thanks guys. Thank you.

Dave Vellante

>> Good stuff.

Rebecca Knight

>> I'm Rebecca Knight for Dave Vellante. Stay tuned for more of theCUBE's live coverage at the Snowflake Summit after a quick break. You're watching theCUBE, the leader in enterprise tech news and analysis.

Mayank Upadhyay, Snowflake & Jason Merrick, Tenable

search

Dave Vellante

>> Here we go.

Rebecca Knight

>> Good afternoon everyone and welcome back to theCUBE's Live coverage of the Snowflake Summit here in San Francisco. I'm your host, Rebecca Knight, alongside Dave Vellante, co-founder of theCUBE. We've got a new guest for the show. I'd like to welcome Jason Merrick, Senior Vice President of Product at Tenable. Welcome, Jason.

Jason Merrick

>> Thank you very much for having me.

Rebecca Knight

>> And Mayank Upadhyay, chief security and trust officer at Snowflake. Thanks so much for coming on.

Mayank Upadhyay

>> Thank you, Rebecca. Thank you, Dave. Real pleasure to be here.

Rebecca Knight

>> Yes.

Dave Vellante

>> Good to have you.

Rebecca Knight

>> So Tenable, of course, is a cybersecurity company. Why don't you start by walking through what kinds of data challenges you were facing as an organization before adopting Snowflake and how it has helped you face those challenges?

Jason Merrick

>> Yeah, absolutely. So Tenable focuses on exposure management. So if you think about the digital assets that companies are responsible for, what Tenable has focused on is helping build an inventory. Sounds very simple, but having an understanding of what cloud assets, IT infrastructure, web services, identities. Oh, AI is also part of that attack surface. So how do you build that inventory and then analyzing that inventory, looking for toxic combinations, misconfigurations and vulnerabilities? And one of the challenges we had as we were going through and acquiring, I think we acquired seven or eight companies in nine years. So being able to bring all that information together and exposure management is about bringing context, having an understanding of the asset, the relationship within the environment, and then being able to help prioritize that for an organization. What are the critical few things? So we had a choice. The choice was, do we build our own data lake? Do we build our own data pipeline? Or do we find a partner that is the best in the industry of doing that? And so we partnered with Snowflake as part of the Tenable One basically data infrastructure.

Dave Vellante

>> So it is essentially your security data lake, if I could use that term.

Jason Merrick

>> Yeah. It's where we feed all of our information and that's where we do our analytics, that's where we do our prioritization and help our customers.

Dave Vellante

>> You used the term, Jason, exposure management. Is that different from posture management or is it just good marketing?

Jason Merrick

>> No, actually it is a program. Exposure management is not a product. If you think about where our roots are as Tenable, it was vulnerability management. So being able to scan servers, look for vulnerabilities and then help prioritize those. Exposure management is looking at the bigger picture. It's not just looking at traditional IT infrastructure. It's looking at cloud, which is CNAPP, which has a component of posture management. Exposures are about being able to link together the asset itself, regardless of what it is, could be a cloud, could be an identity, but lacing it together context within the organization, looking at the configuration, looking at it doesn't have known CVEs on it? Does that have multiple CVEs? So being able to go and highlight that and then be able to reduce that risk. And in this world of Mythos and frontier based models, it accelerates it. No longer can humans keep up with machines, you've got to be able to orchestrate that agentically.

Rebecca Knight

>> So Mayank, when you're hearing Jason talk about these things and then talk about the before Snowflake and the decision making that went into it, what does it signal to you about how other enterprises are thinking about these very same questions?

Mayank Upadhyay

>> Yeah. No, I think Snowflake is definitely a highly trusted brand, which was one of the things which attracted me to come and work here, especially taking care of security and trust. And it signals to me that they get a lot of value out of putting their data in us, not just from the security and trust aspect, but also on the capabilities we bring for data processing, for building that context Jason was talking about, so then they can serve their customers safely, securely, and super efficiently.

Dave Vellante

>> How-

Mayank Upadhyay

>> It's even more important in the world of AI now. Yeah.

Dave Vellante

>> Well, I wanted to go there. Talk about how AI has changed security. I mean, it used to be we could stop 99% of the attacks and the humans could deal with the other 1%. AI has changed that, moving at machine speeds. Just discuss that and what does it mean from a ... How do you attack that from a data standpoint?

Mayank Upadhyay

>> So back in the day when you build software, you would have different components in the software. As an engineer, you knew that this component was going to call this one API in the next component. Then it was going to munch that data and then call the next API there. It was super predictable. But now in the agentic era, these agents are wiring their sequence automatically and you have no idea how this chain is going to flow. So all of a sudden there's all this new exposure, new data paths which things are going to take, and it's completely unpredictable. It's non-deterministic. So from a security perspective, you kind of have to take care of your basics and then have a lot of observability and then that's how you control this, because it's impossible for you to be able to sort of predict exactly what's going to happen each time you run it.

Dave Vellante

>> So I'm glad you brought up observability because you remember the days of Hadoop. We probably could have done it all in Oracle, but it was too expensive so we created Hadoop, which is dump all the data in Hadoop and we created data swamps. I know it didn't really work out that well. But nonetheless, you needed a lower cost platform to do that. Now, we have agents creating all this data. There'll be massive amounts of data. I think people underestimate the importance of observability. How are you observing agents? I mean, you acquired Observe.

Mayank Upadhyay

>> Yeah, that's right.

Dave Vellante

>> It seems to be a really important part of data and now security.

Mayank Upadhyay

>> Let me talk about that. So there's this whole thing, sprawl of shadow AI where people are bringing up MCP servers all over the place, like metaphorically under their desk. And then when developers want to use them, they go down, developers are frustrated. CISOs are nervous because they don't know who's talking to what and what data is being exfiltrated. So the first thing you want to do is you want to put things in a common choke point, have what's called an MCP gateway and have these agents route through them. So this is the reason we announced our intention to acquire Natoma, which is a company that supports over 100 different SaaS apps. And so we believe this is a way that our customers are going to be able to create this paved road for all of their developers. So when they want to talk with AI to different SaaS services, super easy for them and their CISOs can sort of have that observability we're talking about.

Dave Vellante

>> Do you have a perspective on that, Jason? I mean-

Jason Merrick

>> It's interesting. We're in unprecedented times. So you've got every single company feels they're behind. So AI adoption is just rampant and every board member saying, "We got to move as fast as possible with AI," and security is an afterthought. So you've got all this adoption that's happening. And then you've also got these new frontier models that are able to go through and find vulnerabilities. I mean, you look at what Mythos has done, it has really changed the game of vulnerability. So the CISO's caught in the middle. What's winding up happening is Mythos was able to detect 271 vulnerabilities in Firefox. The previous model, four or five. So massive number of new vulnerabilities are going to have to deal with. At the end of the day, the customer still has to deal with that. So CISOs are struggling with, well, how do I handle the employee use case where employees can go and create agents? I can go into Copilot. I can create an agent that touches sensitive data. I'm an employee. I'm not a security person. I may not know how to configure it. So you've got that risk around governance, visibility, observability. And then on the vulnerability side, how are you going to deal with this vulnerability tsunami that's going to happen that you're going to have to go and deal with? And it's no longer can patching in seven days work, you've got to do it in real time.

Dave Vellante

>> So when I talk to security pros, because Mythos obviously created a lot of awareness across the industry, governments, et cetera. They say not only does it expose vulnerabilities, but it lays out the tech path, and that's what's scaring people mostly. Now I know it's not just Anthropic with Mythos. Every major frontier model vendor has similar capabilities. I've seen Oracle responded by saying basically upgrade to autonomous. I saw Red Hat and IBM came out with sort of this open source approach that they're using, both of them kind of monetizing. Have you guys changed the way in which you're advising customers to secure platforms, the protocols, patching sequences? Maybe it's different in Snowflake, but what's changed in terms of your advice to customers as a result of these new capabilities?

Mayank Upadhyay

>> Look, definitely the way things used to work before in the industry has to change. You have to be continuously scanning and continuously remediating. You don't have the option to wait on that. There's a stat I saw where the time to exploit from when a CVE was released used to be of the order of days and now they're saying it's down to an hour or even half an hour. So you don't have time, right? So when you find these vulnerabilities, you almost have to have the AI immediately generate a patch and then have somebody quickly review it, have a human in the loop, especially if it's complicated and then get it out the door. So I think the rate at which software is patched is going to increase. It's probably going to be happening almost on a daily, if not a weekly basis. It's no longer going to be biweekly or monthly or whatever. I think things have to go out continuously. There are many companies in the industry which we're already doing that. I think everybody else is going to have to catch up to that.

Dave Vellante

>> So that speed of exfiltration that Mayank just mentioned, it's not your swim lane necessarily, Jason, other than you are part of defense in depth. How do you use data to improve what you do best?

Jason Merrick

>> So we're moving to a model from detection to preemption. We have to be able to help organizations understand the digital assets they're responsible for across the attack surface, then look for vulnerabilities, look for misconfigurations, look for toxic combinations, help with the orchestration and workflow necessary. And then we just launched our agentic framework called Hexa. And Hexa's focus is helping with that AI remediation automation capability, and validation is the critical other component that validate that the remediation has happened, the patch solves the issue.

Rebecca Knight

>> But in terms of helping the organizations understand there are still so many blind spots to that, Jason. This is objectively terrifying what you're describing here.

Jason Merrick

>> Yeah. Well, I think there's a multitude of things that we're helping organizations with, particularly the CISOs. I think first and foremost, it starts with an AI governance board. We've always recommended you've got to have a fair use policy and employee policy to be able to go through and say, "Hey, what can you do?" I think that there's also some awareness that needs to take place. Shadow AI is a real thing. Tenable helps organizations. We have a capability called AI Aware. So we can actually find shadow AI in your environment, whether that's in the cloud, whether it's in a web service, whether it's externally facing, whether it's in your internal infrastructure or your endpoints, be able to give you visibility into that. And then from there, it's about helping with actionability. How do we take that information? How do we help solve that issue, mitigate risk for the organization? And then at the end of the day, the CISO really has three personas. They have to manage up to the board, to executive management, and they have to do that at the right altitude and they have to have confidence in the information. So we help be able to provide that in almost, I don't mean to oversimplify, but a letter grade, a traffic signal. Not just a list of vulnerabilities that they have to be concerned about. But then their second persona is how do they work with their peer groups? How do they go to the engineering team, the operations organization, the IT team? There, they have to deal with accurate information. They have to know, yes, this is the issue, this is why the remediation needs to take place. And in some cases, it may impact business. May take a system down for a little bit. And then the third piece is managing their own teams. Unfortunately, security professionals, we lack the talent. So you've got to be able to do more with unfortunately the team that you have. So how do we automate that? How do we get the practitioner out of the drudgery? And these are the three things that Tenable really focuses on for organizations.

Dave Vellante

>> How is the structure of defense, from an organizational standpoint, changing? And what I mean is in the cloud era, way back in the cloud era, cloud was like the first line of defense, we had the shared responsibility model. But then developers were being asked to secure infrastructure, shift left. And then you obviously had the SecOps team, you had IT, you had audit, it's sort of the last line of defense. The shared responsibility model with agentic is now changing. So how are organizations or how should organizations respond?

Mayank Upadhyay

>> So I'll say two things. So one of the things we did at Snowflake is we brought both the IT side of the house where our traditional security team, the security practitioners are sided, as well as the security engineering side of the house together because I think the boundaries are getting blurred. First of all, the vendor landscape may not evolve as fast as the threats are, so many things have to be built in-house. So that's one of the reasons you kind of want to bring these teams together. Also, I think AI is a great neutralizer where maybe if you were not a software engineer before, now you can use an agent and you can develop exactly what you need. So the first thing is the organization needs to step back and take a look at what their security or structure is like and see how these two organizations can be brought more close together. The second thing I'd say is Jason was talking about not having enough talent where even if you wanted to scale, there's no way, even if you had the people, you wouldn't be able to scale at the speed at which AI might be attacking you. So the way to do that is through automation. You have to use automation and AI to fight AI. So that is the other big thing security teams have to be looking at right now. It's just a no-brainer.

Rebecca Knight

>> Think like the bad guys.

Mayank Upadhyay

>> Exactly, exactly.

Dave Vellante

>> Well, you have these new roles emerging, the forward deploy AI engineer, which is, now add that in, that individual is going to have to be thinking about security. I think you mentioned these, you can't just bolt it on as an afterthought, right?

Mayank Upadhyay

>> Yeah.

Jason Merrick

>> Yeah. And we're seeing a trend that a lot of organizations are bringing the security team together where prior it was the identity team, the cloud security team, the vulnerability management team, the endpoint security team. We're actually seeing organizations bring those together. Exposure management is one of the titles that they're actually wrapping around those organizations. So we're actually seeing large Fortune 500 companies actually building out exposure management program, which kind of handles what I'll call pre-breach and then detection response.

Dave Vellante

>> And I guess as you just described that, those specialized roles, it made sense at the time because they had to go deep into their own domain, but then what you got was you got data stove pipes.

Jason Merrick

>> Yeah.

Dave Vellante

>> I mean, there was these islands of specialization which now have to come together.

Jason Merrick

>> Yeah.

Mayank Upadhyay

>> And we spoke about a lot of these new things, changes have to be made, but a lot of the basics still have to be taken care of because fundamentally this is all built on a foundation of data. So all the hygiene around making sure you're protecting your data, you've got the right controls and governance around it, whether it's data encryption, the right RBACs, you name it. All those things have to be in place because if that foundation is not strong, agents are going to exploit it.

Dave Vellante

>> Okay. So the what doesn't change, but does the how you do it change?

Mayank Upadhyay

>> Yeah. Yeah, absolutely. In fact, you have to use AI to process these things. You have to use AI to maybe ... I'll give you an example. A person who works at a company a very long time just accumulates more and more permissions to do things. Now in their job today, they may not be doing all of that. But the moment they get an agent to do something, the agent will explore all possible paths. So if you don't want any sort of incidental damage to happen from that, you better clean up that person's permissions. Not just that, when the person sets an agent to work on their behalf, you better give them the right construct so the agent has its own delegated identity and it says you're allowed to do exactly this one thing and nothing more than that. An agent is a little bit like an intern. You give them a credit card and say, "Go buy a printer from me," and they come back with a massive refrigerator and you're like, "What just happened?"

Dave Vellante

>> BMW.

Mayank Upadhyay

>> Exactly. Or a BMW. So those are the sorts of things enterprises have to grapple with. And it's really, really important what platforms you choose and the platforms have all the right constructs for you to be able to do these things.

Rebecca Knight

>> So a lot of enterprise security teams though are overwhelmed by risk and they're under-resourced too. So what is your advice on where they should focus their energy first? Let's start with you, Jason.

Jason Merrick

>> I keep coming back to this, you have to have an understanding of the attack surface you're responsible for. So being able to build that inventory and understanding of identities as that great equalizer that Mayank talked about, you've got to make certain you've got a clean hygienic directory that has the right entitlements. But understanding that inventory, understanding the risk of that inventory and then putting together a program because you have to now move at machine speed. You have to combat AI with AI and reduce risk. I think that's the change that's happening in the industry right now.

Mayank Upadhyay

>> I mean, it's a fascinating world right now. If you look at what is AI security all about, you can break it down into three parts. At the core of it, you've got the LLMs, you want to make sure the LLMs don't get attacked. So last year, a lot of people worked on prompt injection, and so there's products around that. There's really cool stuff you can do today. Outside of that, there's things like we spoke about identity, MCP governance, things which are part of the next layer around the LLM, as the LLM interacts with the external world. And outside of that, you've got your data, which is where you've got to make sure you have the right governance and the right foundation from the beginning. So that's really how you break it down. You got to take care of all three layers.

Rebecca Knight

>> Excellent. Well, Mayank and Jason, thank you both so much. A really interesting conversation.

Dave Vellante

>> Thank you for coming on.

Mayank Upadhyay

>> Thanks so much for having us. Yeah.

Jason Merrick

>> Thanks guys. Thank you.

Dave Vellante

>> Good stuff.

Rebecca Knight

>> I'm Rebecca Knight for Dave Vellante. Stay tuned for more of theCUBE's live coverage at the Snowflake Summit after a quick break. You're watching theCUBE, the leader in enterprise tech news and analysis.