Jennifer Strong: 
Enterprises have been slogging through the AI adoption journey…  but cybersecurity threat actors have been too… and as the time from vulnerability discovery to breach - effectively goes to zero, the new cybersecurity playbook is all about how fast you can recover. 

I’m Jennifer Strong, and this is SHIFT. 

[THEME]

This episode - it's the latest installment of our oral history project.


Anneka Gupta  (00:00):
Rubrik is a cybersecurity company. We focus on security and cyber resilience for organizations. So helping organizations bounce back after they've been hit with cyber attacks. And we focus on helping organizations secure their AI agents so that they feel more comfortable deploying AI agents at scale. Hi, I am Annika. I'm the Chief Product Officer at Rubrik. A lot of cybersecurity leaders, they realize that as their organizations are rolling out AI agents, that there's a huge amount of risk, but some of them have kind of thrown their hands up in the air and they said, well, I, there's nothing I can do. I just, I'm getting this mandate put at me that I need to, uh, make sure that AI agents are rolled out everywhere. I don't have the ability, the time, the people to go figure out what I can do to truly manage the risk.

Anneka Gupta  (00:49):
And I think that's a big problem. Mm-hmm <affirmative>. I think that a lot of organizations need to recognize like, Hey, I need to be doing something ahead of time before it's too late, before I have AI agents running everywhere and I have no way to, to manage and control that. A great analogy that I feel like has resonated is that when Public Cloud first rolled out, a lot of organizations were like, Hey, you know, I have, my developers are spinning up things in the cloud. I have no control over it. They didn't put in place all of the mechanisms to monitor environments to ensure their cloud security was really strong. And then they've had to do that after the fact as they've had disruptions to their environment as they've had cyber attacks. And what I'm hopeful for for AI agents is that people will learn from that mistake and will ask themselves the question, what can I do today? What should I be doing today to lay the foundation so that I can roll out AI agents safely and securely for my organization?

Anneka Gupta  (01:49):
Agent Rewind is a product that we rolled out in August of last year, and the origin story of Agent Rewind is one of our founding engineers came up to me, I think in like June and was like, Hey, we're rolling out AI agents internally, and AI agents can do so much, so much more than a chat bot. The reality is, is that we're rolling out all these agents and they are going to make a mistake. Something is gonna go wrong, they're gonna delete something. And isn't this a great opportunity for Rubrik? Isn't this the reason ano yet? Another reason for why people will need recovery is to be able to recover from agent mistakes. And after that conversation, I was so energized because it made complete sense to me that this was going to be a problem, may not be a problem of today, but it was going to be an imminent problem as people rolled out agents. So we quickly got a Tiger team together and in eight weeks launched Agent Rewind, pulling together different pieces of the product, but building a completely new experience that actually started with an agent inventory, looking at all the actions they were taking, and then connecting it to a very specific recovery point or recovery plan to be able to recover.

Anneka Gupta  (02:59):
If an agent makes a mistake and and deletes or changes something you didn't want it to do. When Philanthropic first announced Project Glass Wing, we were really excited about it and we wanted to be a part of it because we knew that Methos was completely changing the cybersecurity landscape. And we started getting questions from clients who are asking us, Hey, like what are you doing, um, with Methos? Like, how are you securing your software? And so we actually started playing with some of the other AI models to look at opportunities for where we could harden our own software to prepare for methos. And then when we got into, when we heard we were getting into the next wave of Project Glasswing, we already had a whole project plan in place and a team that was ready to jump in and really start experimenting and figuring out how we could get some valuable insights and then, you know, patch things that were, were in our code base or share some learnings for some of the open source software, which a lot of other companies have done. The challenge is though, is that open source software has a lot of vulnerabilities in it, and that is means that it's not just a vulnerability. If we find a vulnerability in open source, it's not just for us, it's for the millions of other people that are using this as well. And so I think that's the place where collectively as companies like all All Together, we have to figure out how do we help the open source community patch these, these software because that is going to have super high impact.

[MIDROLL]

Anneka Gupta  (06:00):
It's kind of crazy. I wouldn't have ever predicted that this is where I would've ended up, but I think what was really exciting to me about Ad Tech was that at that time in the 2000 tens, it was really like the renaissance for advertising and everyone was trying to figure out how to advertise online. The company I was at, LiveRamp was in the data space. So we were helping connect data to help people advertise better on the internet. What was super interesting as I went on that journey, I realized that data is such an important asset for companies and we were begging companies to give us more of their data, our customers, to give us more of their data so we could help deliver them better outcomes. When I started looking for what was next, I really wanted to be at a company that was in a highly innovative space because I loved that about my early journey at LiveRamp.

Anneka Gupta  (06:48):
And I found that there was this through line with data because unlike LiveRamp, like Rubrik was in a place where because we were protecting organization's most valuable data, they were giving us their data asset. And that meant that we could build so many more products on top of critical data and not have to be like begging people to give us more data. They had to do it as part of the core product we had. And I just felt like there was so much potential to expand the portfolio way beyond what we were doing five years ago.

Anneka Gupta  (07:20):
I would say that since 2021 when there were very publicized, big cyber attacks, for instance, on Colonial Pipeline, that really changed the nature of the board conversations that were happening within our customers organizations. So that's when we started to see a much bigger pull because boards were saying, Hey, what are we gonna do in a ransomware attack? Are we gonna be able to recover? And that was getting pushed down to the people that work with Rubrik each and every day to, to showcase the solution. So that was actually a really big moment for us because we were able to, uh, attach ourselves to that tailwind. And now all, all organizations, all markets, all investors are also looking to companies and saying, Hey, how are you gonna be a lot more efficient with ai? So there's this massive push to say, Hey, how can we show AI productivity across our organization? At the same time, AI is opening up a huge new set of risks because we're going from human attackers to now AI attackers. And then when you start saying, Hey, I'm also gonna open up my organization and have all of these AI agents that have a massive amount of access to the different applications and data within my enterprise. When you layer that on top, an attacker's also trying to go after and compromise agents, all of a sudden this problem around cybersecurity and risk is, is now a hundred x what it used to be.

Anneka Gupta  (08:48):
Our core business of cyber resilience and cyber recovery assumes breach. We assume that a customer has already been breached. How do we help them prepare and be able to recover quickly so they minimize their overall business downtime? Mm-hmm <affirmative>. That part of the business is only more relevant now than it's ever been. And now we are also using AI agents and that is a really new innovation that allows more reasoning and adaptability to cyber scenario and allows organizations to recover faster, which they need to because AI is creating machine speed attacks. On the other side of the equation, you have all these AI agents that are running wild. And when I talk to customers, their biggest problem is I don't have any visibility into what my agents are doing and I have no way to control them because I have no visibility. And so we have another product suite called Rubrik Agent Cloud, where we focus on building a holistic inventory of all the AI agents that sit within an organization building a control layer where we actually in at runtime are looking at all of the traffic happening between the agents and the models. And you can set up policies like, I don't want my agent to send emails to my customers. I don't want my agent to be giving financial advice. And we will enable those policies to be monitored and enforced in real time using an AI agent. And then we help with the recovery too, in case your AI agents go awry. So our whole portfolio is really focused around two problems of assuming breach and assuming agent overreach.


Anneka Gupta  (10:25):
So it depends on what's going wrong. So an AI agent could be making just a benign mistake. That mistake could still have massive repercussions. So there's a lot of publicized examples of ai, an AI agent deleting a big piece of production infrastructure or deleting inboxes. So that's something that can happen and all of a sudden you realize, wow, okay, a core application isn't running anymore or I'm not able to log into my email. Those are, that's when you might catch it and then you'd have to figure out how to recover quickly. And that's where Arik comes in. When you look at a cyber event, there's a lot of different kinds of scenarios that can happen. So one is typically what the cyber attackers are still getting into organizations the same old way that they've been getting into organization, which is compromising credentials. But now they're using AI to look a lot more like a legitimate email with the legitimate link that you click on.

Anneka Gupta  (11:19):
Then they come into your organization and they're going to start searching around for what are the crown jewels? How do I get to the most valuable data? Now, if they can compromise an AI agent in that process, that's gonna accelerate their ability to get to the crown jewels because a lot of AI agents have a lot of access and then they're gonna move laterally. They're going to find those crown jewels, they might exfiltrate data and threaten to put on the dark web. They might delete or create some kind of disruption within your production environments. And you know, at some point in this process you get an alert that something has gone wrong. This is

Anneka Gupta  (11:52):
Where we, we are trying to help pull together all the pieces and do a lot of the investigative work on behalf of the customer so that they can quickly respond and recover to this scenario. So we'll be able to say, Hey, here are all the agents that are running. We'll be able to show like where there was a change from, um, the baseline where there was some high risk activity that was taking place, and then be able to show here's actually where you need to recover. And alongside of that we'll also show that for your data, for your identity systems, so that you can holistically across your data identity and agents cleanly recover.

Anneka Gupta  (12:34):
I do think that there is a shift that's happening. I do think that organizations realize that if they get hit it's or when they get hit, it is better to be transparent with customers and externally about what's happening and share their learnings about how they're responding and remediating as the event is happening versus six months later. So that I feel like is becoming a norm and organizations are being judged on how well they're transparently responding to this. But I still think that there's a lot of hesitance to share, and I understand. I understand where that hesitance is coming from, but it's still there. And it's not fully legitimized to say, Hey, this is a breach that happened. Here's what's going on here was what the impact, here's what we did about it. Here are the learnings for everyone else.  

[CREDITS]