Skip to main content

The Titans of Trust: Powering the Open Security Ecosystem

Summary

  • This video features a panel with leaders from Wiz (now part of Google), Palo Alto Networks, and Zscaler discussing how an open security ecosystem enables organizations to defend AI-native environments, where the security perimeter now spans from infrastructure all the way to the AI model layer.
  • Panelists explore new challenges unique to AI-native environments including tracking provenance and ownership of AI-generated components, identifying which AI agents are active in a cloud environment, and covering blind spots that siloed security tools miss.
  • The discussion covers how Databricks is building an open security ecosystem with ecosystem partners to enable machine-speed threat response, cross-platform integration standards, and evolving governance roles for CISOs as AI autonomy increases.

The Titans of Trust: Powering the Open Security Ecosystem

Watch: The Titans of Trust: Powering the Open Security Ecosystem
As organizations transition from cloud-native to AI-native, the traditional boundaries of security have vanished. To defend at scale, security operations must be as interconnected as the systems they protect.
Join industry leaders from Wiz, Palo Alto Networks, and CrowdStrike for a strategic dialogue on the convergence of data intelligence, AI, and collective defense.
The discussion will dive into how visibility now spans from the infrastructure layer to the core of the AI model. We will explore how the Open Security Ecosystem creates a unified fabric for developing and deploying AI agents safely—allowing teams to neutralize complex attack paths that siloed tools miss.
this video offers a rare look at how industry leaders are partnering to ensure that the velocity of AI innovation is matched by the resilience of its defense.
Talk By: Ashish Kathapurkar, Sr. Director, Databricks ; Jiong Liu, VP of Product Marketing, Wiz ; Spencer Thellmann, Principal Product Manager, Palo Alto Networks ; Vineet Parmar, Zscaler ;

Chapters

FAQs

What is the Databricks open security ecosystem?

The open security ecosystem is an initiative where Databricks partners with leading security vendors — including Wiz, Palo Alto Networks, and Zscaler — to build a unified fabric for developing and deploying AI agents safely. This video explains that security in AI-native environments cannot be done in isolation, and cross-platform integrations are essential to cover the full attack surface.

How does AI change the security perimeter for cloud-native organizations?

This video's panelists explain that as organizations move from cloud-native to AI-native architectures, the traditional security boundary expands to include the infrastructure layer all the way up to the AI model itself. New challenges include tracking provenance of AI-generated code components and identifying which AI agents are active in a cloud environment and what they are building.

How do security teams build trust in autonomous AI agents?

Building trust in autonomous AI agents for security operations requires guardrails, defined risk tolerance levels, and cross-vendor integration standards that allow consistent policy enforcement. This video covers how security teams can design governance frameworks that account for both external adversary threats and insider misuse scenarios in AI-native environments.

What is Prisma from Palo Alto Networks?

Prisma is Palo Alto Networks' security product for protecting applications, models, and agents, described in this video as their fastest-growing product. The panel discusses how it integrates with the broader open security ecosystem to provide cross-platform visibility and policy enforcement for AI-native workloads.

Full transcript

[00:09] Welcome folks. Welcome back. Uh those who have taken a break, welcome back. Those who are here for the first time, welcome to the session. U this is a different session. We are going to have some of our ecosystem partners join me today to talk about open eco security
[00:22] ecosystem. Throughout the day, those who have been attending these sessions, um, you've heard from security practitioners and leaders across different organizations on the challenges of
[00:34] security. What I heard through those talks was one un undeniable truth is security cannot be done in silo. And so here at data bricks, we are building an
[00:46] open e security ecosystem. And I'm pleased to welcome today on stage to join me leaders from whiz now part of Google Palto Networks and Zcaler
[01:00] to talk about how we are building this open e security ecosystem. So please give a round of applause to um Gian from whiz.
[01:22] >> Thank you. Uh we have Venit from Zscaler
[01:36] Thank you for joining me here today folks. Um before we get started, I think it would be great if uh you can take a minute or two to just uh introduce yourself, what your role and scope is uh at your company and just give a brief
[01:48] background. >> Yeah. Uh I guess I will go first. Hi everyone. I'm Junglu. I lead up the product marketing function at Whiz. Uh where we're in charge of product go to market strategy. Been very fortunate to
[02:01] be part of Whiz for a bit over four years. So definitely have seen a lot of growth from when we were just 200 people to then 3,000 people and now part of Google. And so we're incredibly excited
[02:14] actually to be joining Google at the exact moment that um you know everyone is now adopting AI and building AI applications and really excited to be amongst the broader set of e the
[02:28] security ecosystem as we all endeavor to make our spaces more secure and hardened particularly in the AI era. >> Venit. >> Hi good afternoon everyone. Venit Parmar. Um I lead uh a new role at
[02:42] Zcaler. So I head up our AI partnerships. Uh I've been at Zscaler for two years. Prior to that I led our partnerships with uh both AWS and Microsoft. So um excited to be here and thanks for the invitation. Spencer.
[02:56] >> Hi everyone. My name is Spencer Telman. I'm a principal product manager at Polit Networks and I help our customers to secure their apps, models and agents with Prisma which is our fastest growing product of all time now. Uh so I'm really happy to be here and excited
[03:09] about the announcements that we put out this morning. >> Thank you Spencer. So we have a diverse set of panelists here today. Uh but before we kick things off, there's one thing which is very important here to to see how we can shape things up um to
[03:21] fight some of the modern day attacks, it is becoming very critical to sort of unify security across infrastructure um identity uh data AI and different assets together. And so to get to that
[03:35] unification, the first question that we need to sort of look back and answer is what are the blind spots here? And there are certain blind spots that we are not aware of. And so I'm going to open this
[03:48] up with Jong to understand whiz has a u unification strategy across on-prem cloud uh different assets across data AI. We announced some integrations as well. So what in your mind are some of
[04:01] the key blind spots that security practitioners need to be aware of especially these days when AI is so uh prevalent everywhere? >> Yeah. So I think even just um as we think about how development has shifted
[04:14] very dramatically in over the last 6 months um when you think about it right a lot of organizations are increasingly building now with coding agents. They are adopting new components into the
[04:26] software development pipeline. their agents now have access to lots of data and the ability to actually take actions on their behalf. And we're also leveraging not only our custom code
[04:38] here, but also third-party components as well. And all of this now gets shipped through our software delivery pipeline faster and faster. And beyond just even our traditional product engineering,
[04:51] security sort of trifecta where we have security reviews, right? We have a lot of folks in organizations that can vibe code and what we're seeing actually increasingly amongst our customers is the prevalence of just shadow AI things
[05:06] that are appearing from different teams teams that are also looking at um building agents across not just the cloud infrastructure but of course partners like data bricks that we have um and they're also looking at no code
[05:18] options right how do I potentially leverage let's say like a Salesforce agent force to help me build and so we're seeing Actually just the prevalence of blind spots across not only the technology components that we
[05:31] use but where the provenence comes from and even ownership right becomes actually one of the trickiest blind spots for security teams as we now evolve beyond just the the core of who
[05:43] we traditionally considered builders. >> Thank you. Uh
[05:55] what are some of the key blind spots that you when you talk to your customers or when you talk to sock analysts that you see on the AI side that we should be aware of? >> Yes, I think there's a couple different ways to think about this. Uh, we talk about shadow AI, but really I think that
[06:08] companies have multiple different shadow AI problems within that, right? If you think about the history of securing AI, it's really as a discipline probably about three years old and it's had about four different turns in my view. The
[06:20] first was uh companies were concerned about their employees dropping their source code or things like patient data or some other protective data patterns into someone else's chatbot. Right? So that was like the first frontier. The
[06:33] second was uh inevitably enterprises start building their own chat bots that are exposed to their users in their apps or on their website and then you're concerned about people jailbreaking your chatbot or you accidentally saying something off-brand back to an end user.
[06:46] So that was like the second turn. And then the third was uh agents both procode built in the public cloud or or no code things like agent force. And then the fourth major turn that we're seeing now is the endpoint where every
[06:58] single day I get asked about all these different kinds of endpoint agents. You know where companies want to know who's using what? What are they doing? Which MCP servers are they calling? How can they scope this down? So if anything, you really have four shadow AI problems. You have what are your employees using?
[07:11] What are you building? Which agents are you putting together in the cloud and in these noode platforms? And then finally at the endpoint, what are your employees doing on their laptops with agents? Each of those has their own discovery problems.
[07:23] >> Awesome. Uh so my take here uh folks is obviously there's an AI stack here whether that starts from infra all the way to apps or frontier models and uh whatnot. Um I'm going to ask this
[07:35] question to since you focus on partnerships and ecosystem here. How do you think about building this ecosystem around AI u especially now? Um and and how do you how should we think about
[07:47] building this whole ecosystem to make sure that we eliminate these blind spots? >> Great. So, you know, when we talk to our customers today, I mean, we're seeing a tremendous amount of AI sprawl in the enterprise. Um
[08:00] Zcaler's latest threat labs reports, you know, what we saw last year was almost 990 billion AI ML transactions in the enterprise today. Um, and so when we talked to security leaders today, the biggest thing that they're concerned
[08:12] about is, hey, how do I get visibility into the data and how do I secure the data in this world of agents? Um, we talked about AI sprawl. Um, it was it was and so we're seeing a lot about that. Um, we're also seeing, you know, a
[08:25] lot of concern from customers around I'll call it agent chaining. Um, so in the past it was really about, hey, how do I secure the vendors that I'm working? Now it's in this world of ATA and agentic exchanges. Um, we now have
[08:37] agents talking to agents which are talking downstream into open source connectors. And so how do we think about securing that? Um, and then as agents go from answering questions to taking action that increases a significant
[08:49] amount of um, um, landscape around threat. Um, and so for us the blind spots really sit I'll call it at the seams. It sits across identity. It sits across the devices. It sits across the network and it sits across the data. Um
[09:01] and so the way that I think about it or the way that we think about it is that to help with the blind spots um it's really about how do we share um televatory and visibility into what's happening across the AI stack because
[09:14] there's no single vendor that works across that entire AI stack. You talked about infrastructure all the way to the applications. And so if we look at the partnerships that we're building, um it really is from the infra lever to how do we think about data and AI partnerships
[09:27] like we have with u data bricks to frontier models all the way up the application. So that's really how we think about it. >> Awesome.
[09:40] on is as we think of these blind spots and go towards more building an ecosystem and start to rely more on solutions to help us eliminate these blind spots. What does trust really mean from a buyer perspective and from
[09:53] security practitioners perspective? And you touched based on that a little bit is telemetry and visibility. Um but I want to double click with each one of you here is what what should we be looking at from what does trust really
[10:06] mean here uh in in when we build this whole ecosystem around AI infra identity how should we thinking about that is it certain metrics that we should be measuring ourselves against is it outcomes is it uh how should we think
[10:18] about this >> yeah I think it's a really interesting question and I I want to just kind of add on first which is I think because of everything that's happening with AI,
[10:30] right? It's very clear that every organization has to now open up their ecosystem in order to share telemetry wherever possible. And this actually makes building security products now
[10:42] like there's been no more exciting time. And if we think about like what is all of the telemetry that is available to us, we really as security owners, we have a a giant home field advantage,
[10:55] right? Because we understand our applications. We understand our teams. We understand what's critical to our organization. It's just historically all of this context has lived in different places. So when we think about trust,
[11:08] it's really how do we get really high quality context together that really establishes a concrete understanding of this is what an application is, right? This is how it's comprised from the
[11:20] cloud infrastructure to the data and access to the identity to the models and guardrails to the actual application. And then how do we map that entire lineage back through the pipeline it was built to the code repository to the
[11:33] owner and then we can think about layering on agents right so if we give agents now access to this context do they actually now inform us with precise actions that allow us to proactively
[11:47] harden this environment or monitor the environment in a way that we are confident in. So we see trust more as um not only do we trust our understanding of the inventory that we have but also
[12:00] are we able to now increasingly build trust in the agents that are helping to secure the environment. >> Thank you Spencer. Anything to add to this in terms of any metrics that we should be looking at? Yeah. So, uh I
[12:13] think especially in the context of AI, what I've learned now working in the space for three years on on AI security is that like the utility of your product is a function of the degree to which you embed yourself in the tools that people
[12:25] use. That is a really long sentence, but what I mean is that you cannot be insular with AI, right? You have to open the sort of stack up that that you serve to end users because ultimately you want to make sure that you integrate with
[12:37] whatever it is that someone uses. And the truth about that is that that means that your partnership teams also have to move really quickly because some of the endpoint agent stuff that we're talking about right now wasn't even a thing like nine or 10 months ago. And so that's
[12:49] kind of how we think about this is that like we always want to make sure that we have the capability to integrate deeply with whoever comes about. And a lot of that boils down to just making sure that everything's an API. Uh but on the trust side, um something that I think about is
[13:03] context, right? inevitably we're going to have to get to this sort of future state wherein we can judge the intention of an agent before a decision is made and I think uh the only way to do that like at a degree or at a probability
[13:17] that's actually meaningful is to gather context from the things around the agent which will involve a degree of trust uh across parties in that case. So that's kind of roughly how we think about that. M so it's interesting you say this uh if
[13:29] you folks heard the keynote today Ali also mentioned three things context control and choice and so Venit my question to you is how do you think ecosystem across these three pillars if you will right how do you get the right
[13:41] context when you build open ecosystem uh how do you control the different facets of this and what's the choice choice is obvious customers have their own choice to bring in their own solutions and work
[13:53] with different vendors >> but how do you focus on context and control. >> Yep. So I I think of it in sort of in three ways. One is uh from a trust perspective. Um a how do I think about
[14:06] trusting the data is it unified and can it be governed? Um the second thing that we look at is from a from a tooling perspective and and talk you talked about a little bit around hey how do I ensure that the tools that I use to
[14:19] query the data um it's not a closed ecosystem but I have choice around that. Um and then the third one is how do I think about trusting my AI in the context of policies that I need to be able to set up. And so throughout the conference today we heard a lot about
[14:31] you know thousands and thousands of signals that customers are seeing today. Uh data from CSPs, data from vendors um from an open an architecture and trust perspective. We really think about a um
[14:43] how do I enable enterprises to ensure that the data they have is single source open framework. Um I can get my security signals in there. I can get my um IT signals in there or I can get my uh
[14:56] business signals in there. From there, I want to ensure that we talk about is we want to ensure that customers um have the right set of tools to query the data um and not locked into a single vendor cuz I think we hear a lot around you
[15:10] know is it the right data is is it is it the correct models etc. Well, if you allow customers to use the right tools to query the data, you're not having conversations around, you know, is it the right data or not. Um and the third one is really around how do we ensure
[15:22] that as you're building agents on top of the on top of the platform um they're they're really starting to drive the right insights and so we think about guardrails both guardrails built into the platform but openness around having
[15:34] uh other vendors being able to integrate into those guardrails um so that the agents are driving a secure motion on on top of that. >> Nice. And so u just to round some of these things up while we focus on trying
[15:46] to eliminate these blind spots or first finding these blind spots trying to eliminate these blind spots and building the trust with uh the right ecosystem here the key question that comes to the mind is how do organizations need to
[15:59] structure themselves whether it's culturally or organizationally to create all this unification so that they can eliminate the right blind spots and be able to move forward with AI adoption here. Um I remember back uh this is
[16:14] early 2000s when I wrote code in embedded systems uh the software deployment cycles were like 6 months long or 9 months long especially when it comes to monolithic writing software
[16:27] that shrunk to now in weeks if not days maybe in hours moving forward and so my question to all of you I'll start with maybe uh Spencer to begin with how should organizations think from a
[16:39] culture perspective ive of setting up their development teams, test organizations, uh go to market teams, product teams. Uh is there a shift happening here? Should there be a shift happening here to to to adapt to the
[16:52] modern times? >> Yes. If we talk about this through the lens of securing AI what I've noticed is that the the org chart is highly non-standard, right? Every company I speak to is totally different. Sometimes it's network security people that are
[17:05] now responsible for securing AI. Other times it's like the engineering org. I see a lot more of that recently. I think eventually sort of security will become engineering discipline, not the other way around. Uh and and what that means
[17:17] is that the person or the team in your company that owns AI security kind of also defines how it's going to be done, right? Because a network practitioner is going to approach that problem totally different than someone who works in
[17:30] engineering who might have a proclivity towards SDKs and APIs and the like, right? So that's the first kind of interesting part. Uh but the second is that I I strongly believe that security again is becoming an engineering discipline whereby like as you're
[17:43] building the thing as you're putting your agent together it has to be embedded kind of then like ideally the the strongest teams that I meet the people building the agent are also responsible for securing it which means
[17:55] doing all of the supply chain and the runtime and the red teaming etc where the edge team owns that. So it's having some really interesting implications on I think how like org charts are structured but ultimately this is like highly non-standard. Every company is
[18:08] slightly different. >> Sure. Um and so we've already seen CISO becoming a part of the board um in several organizations. Um Venit in your view when you talk to several of your
[18:20] customers or your ecosystem partners do you see that role changing uh at board level or even at microorg level where even development teams need to have some element of security knowledge or
[18:32] awareness built into these teams. >> Yeah. So I I'll answer that in the context of you know the conversations we're having as we get questions around sort of security analysts and what the role of that security analyst looks like. Um and so we believe that you know
[18:46] today the analysts need to move away from general analysis of signals to how do we think about them being orchestrators of AI agents. Um and so what I mean by that is uh they got to move away the future is really about
[18:58] moving away from just quering logs to how do they look at um understanding um the the signals that the AI is delivering. How do you validate that in
[19:11] in the framework of what they're working through? Um and then how do you then um automate that so that the responses that come back you're able to automate it in terms of um key um key ways in terms of
[19:24] the threats that are coming in. So that's kind of what we're seeing the the the analysts kind of look at and what we have conversations around is hey how do you build capability around understanding identity um and access um
[19:36] so who has access to what where and why uh and then really ensuring that you start to uplevel your skill set around um understanding how agentic attacks work um because it's very different than how we think about uh uh uh users and
[19:50] threats um and so that's sort of the conversation that we're having um with those type of personas. Yes. Uh Jang, anything from your perspective, especially culturally, right? Because organizations also have to start thinking from a culture perspective. Uh
[20:02] the development cycles have shrunk, uh customers expectations have changed. >> Um and it's going to accelerate further. So culturally, how should we start thinking about these organizational shifts? Yeah, I think in general if you
[20:14] think about even how security teams have traditionally been organized, right? We were very much segmented vertically sort of based on where you were in like the development life cycle, right? So absc
[20:28] right scanning code for code risk. If I'm in DevOps or dev sec ops, I'd think about the pipeline. If I was on the cloud security team, I thought about cloud. If I was in sec ops, I thought about like detection and response. And
[20:41] when we think about now having to all move at machine speed, right, these like silos, they don't work. So we actually must operate horizontally. And that means not only do we have to sort of
[20:53] automate a lot of the manual work and automate a lot of these like handoffs that we used to have. I think the biggest cultural shift actually is really encouraging sort of that endtoend
[21:05] architectural thinking and really thinking like uh about curiosity and encouraging curiosity wherever possible because while AI is changing like the ground beneath our feet like on a daily
[21:17] basis the reality of breaking silos is that it's a fundamentally human it's only a human that can really do that right that's how you reach across teams understand what their teams have
[21:29] historically done and redesign something that works for the future state. So I think actually that sort of human thinking the curiosity is really really critical. >> Uh well that's actually my very good
[21:41] segway to my next question is what are the skills that we should be thinking of because the skill sets obviously needs to be either upleveled or looked at very carefully for both security practitioners or for kids who are coming
[21:54] out of college. what what are the key skills that we should be looking at? I think judgment right is actually one of the most critical skills because if we take you know let's say like a level one
[22:06] sock analyst right historically their job was investigate but now an AI agent will investigate and it will reason over context and so as a level one sock
[22:18] analyst like my job is actually do I trust what the agent told me right it's sort of that like validator and so I think it's a a skill set that you see in that first level but really at every
[22:31] single level and tier above that as well. >> So are you more referring to I'm using the industry terms that I'm keep on hearing is human in the loop versus human over the loop. Uh would you like to use any of that or have your own term
[22:44] here? Uh I I mean I think it's an evolution that we're on right right now we are in a human in the loop state but you can't just only think about the human in the loop right you have to
[22:56] think about what is the broader architecture what is the broader agentic trust journey that we're all on because I think you know many of us have seen this with self-driving cars right for a long time we weren't that trusting of it
[23:10] but over time we built the trust and we're more comfortable with that and you know we all take wayos probably all of the time now. So we can't just think about the current phase, right? But we have to think about how we continue to
[23:22] mature and continue to build trust in the current workflow as well as increasing the fidelity over time.
[23:36] practitioner role itself moving forward? No, I think I think it's kind of the conversation we just had, which is really around moving away from, you know, quering logs to really thinking through um how you're leveraging AI uh
[23:52] to drive insights and then drive uh downstream support around that. >> Yeah. How I think about this is that like the the skills that seem to matter now are like agency and a bias towards creation. uh right like but that also
[24:04] has some like in fascinating risks and you think about this through the lens of endpoint agents right now with endpoint agents anyone can create like I went to school for years for a computer science degree that I'm not sure I need anymore now right um and so I know you know SVPs
[24:18] that create their own prototypes without ever talking to design and engineering and that's amazing right because anyone can create anything now but at the same time you have to think about the risks that this exposes your organization too and it has some interesting nuances too
[24:31] with things like FinnOps even where you've probably seen the news recently about like token economics and how these like strange misaligned incentives within companies are encouraging employees to burn tokens needlessly because they want to be at the top of their internal leaderboards, right? It's
[24:44] this strange kind of phenomena where ultimately like the the mental model through which I think about this is you want to give people like a playground to create within but a playground that's sort of hardened and secure just like one is for kind of children in a in a
[24:57] proverbial sense. Right. Nice. >> Um so that's kind of the lens through which I see that. >> Awesome. So, so folks u so far what we've heard is sort of three big areas. One was how do we
[25:09] essentially to reduce attack surface is what we call in security. Two is what does trust mean in the new age and what are the skills and organizations how should they be thinking about culture as
[25:21] well as organizational alignment internally to move to the new age. Uh thank you for being attentive and please a round of applause to the wonderful panelists here.
[25:36] Thank you. Have a great rest of the summit.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.