Agentic PAM and the Identity Era That Broke IAM

Hey, everyone. Welcome back here to Techstrong TV. My next guest up is Jeff Malnik.

Jeff is the VP GM of Developer and AI over at 1Password. Jeff, welcome to Techstrong TV. How are you?

Hey, I'm doing great. Thanks for having me today. My pleasure.

So Jeff, I always like to let people, give them a little insight into who they're listening to, who we're talking with. I gave them your title, but give them a little bit of the Jeff Malnik story. Yeah, sure.

So, I've been at 1Password for about seven-ish months now, seven or eight months. I joined by way of HashiCorp, where I led the engineering group for the secure products team, which focused on tools like Vault, Boundary, and Radar. Sure.

Before that, held a couple of different roles at various startups and fintech companies. Very cool. Doing the startup thing.

I've kind of done a few startup things myself, actually. But glad to see you here at 1Password. We said GM of Developer and AI.

What exactly is that? Yeah, that's a great question. So my group focuses on the next generation of machine and agentic services that 1Password is providing to secure different types of machine and AI workloads.

So, historically, 1Password hasn't really been known as a tool that goes beyond human use cases. A lot of people out there still consider us a tool that's focused on providing just access to credentials in a vault. The capabilities that we're rolling out previously or already this year, and are going to continue to roll out later this year, are more focused on agentic use cases, machine workloads.

1Password historically hasn't been known as a tool that's served machine and agentic use cases. We've been known as a tool that provides access to credentials that get you access to upstream systems. Right.

We've known this for a long time, but what's interesting about where we're going today is that we're now, especially given the recent news about our acquisition of Epona, we're able to provide access not just to the gateway and to the entry, if you will, to the application, but being able to control and govern the access for the stay within that application as well. And so super excited about what we're building there. Along with serving some of these AI and privileged access capabilities, my team also focuses on our developer tools.

So you can always reach out to me if you have questions about our SDK, our CLI, our environments capability, and a lot of the other developer-facing capabilities that we have. Excellent. So Jeff, we were talking before we got on on camera here.

In my mind, when we look at agentics, we look at AI agents, it represents sort of the next explosion of managing credentials, of managing identities, right? When we moved to the cloud, identity and access management sort of became the-- Well, let me back up. Before we moved to the cloud, like when I was doing security companies, it was the moat and castle.

Yeah. Right? You basically had all your workers within an office.

There was a moat around it, and you had a drawbridge, right, that came down and that controlled ingress and egress identity. Right? And it was a choke point.

When we moved to the cloud, that really became very different and all of a sudden, really, identity became important because you weren't in the castle, you were everywhere. And how were you getting in, and how was one cloud to another? Excuse me.

Then came the age of machine identities. Whether we're talking about IoT devices or even containers within a cloud-native Kubernetes stack, new credentials, new identities, new things we had to manage, accesses that we had to manage. Now we have agents.

And agents are a funny thing because one can make the argument that each agent should have its own unique credentials, its own unique security identity in terms of where it could go, how it could go, when. But so many of these agents are riding the credentials of the people you like. In my case, my agent rides my credentials, right?

And how do we know when that's an agent doing it versus me or another person? This is hard. Let's not kid ourselves.

This is going to make for some hard choices. We've got to have solutions that can handle it. How's 1Password helping with that?

That's a great question. So one of the things that makes agents really challenging is that, you kind of hit the nail on the head. They're machine workloads, but they have a twist.

They have a reasoning capability, and so they behave a lot like a human user would, but they talk like a machine, right? So they have protocols and all the existing protocols that we have for securing machine workloads still apply, but they've exposed some pretty non-trivial gaps in those protocols. One of them is delegation workflows, right?

And we're seeing a rapid acceleration of closing the gap there. Anthropic's recent announcement of supporting enterprise-managed auth, for example, and the IDJAG capabilities that folks like Karl Maginnis and Okta have kind of spearheaded over the last several years. We're leveraging now emerging protocols to go and help close a lot of these gaps.

It's not enough to say that you're going to go through a delegation workflow once, apply a static policy, and hope for the best. Arguably, in the past, that was already broken. But it worked because machine workloads, for better or for worse, were pretty static in how they behaved.

They were predictable. We kind of knew that they would access this service and need access to this table and this database, maybe down to row-level capabilities. We don't have that with agents.

An agent might decide that, "Oh, I need to get access to a different table in that database," or maybe a different database altogether. That means that we have to have systems that are responsive to that dynamic nature of the agent workload itself. We have to build what looks a lot like human types of security around those machine workloads.

So things like approvals workflows, step-up authentication, step-down authentication, background re-authentication, things that allow it to behave more like a human, but at the same time, within the existing protocols that we have for machine workloads. So there's a lot of different technology that's emerging around that right now. 1Password is focused on building out new capabilities to govern the state that that agent might have in an application.

So for example, with our acquisition of Epona, we're going to be more focused on building just-in-time access stories around the credentials that you already store in your vault. So the ability to allow humans, machines, and agent workloads connect to our product and serve just-in-time access capabilities around those different upstream resources. I think a lot of this is starting to converge on what we would term agentic PAM.

That's basically the way of taking existing machine workloads and creating PAM-like capabilities that have traditionally served human use cases. We now have to do this for machines. It needs to be much more auditable.

In the past, I think we focused a lot on session recording, which still applies in the era of agentic PAM. But at the same time, we have to have the number of agents that are actually spinning up at any given point in time as many. And the simply recording a session and then doing analytics on it after the fact or doing investigation on it after the fact might not be practical with the volume of agents that are actually spinning up.

And so we need to have more structured logging, more structured telemetry around what's actually happening in that system, and that's on the other side of that security story, which is an area that we're also focused on. I love it. That's excellent.

Jeff, people I'm sure out here have questions. They want to get more information. A lot of people are just starting to kind of wrap their hands around this because you can't come up with the solutions till you really understand the issues, the problems.

Go to 1Password. What advice do you give people to get smart about this stuff? Yeah.

I think the most important thing to understand is that you wouldn't just walk out onto a street and give a perfect stranger your laptop. It's the same thing when you're dealing with AI agents. One area that we're really focused on securing when it comes to agentic workflows is your laptop.

Today, we exist on millions of endpoints. People have trusted us with millions, billions, I should say, credentials securing those different systems. What's interesting is that up until a couple of years ago, when agents started to proliferate, really the only person who cared about clear text credentials on a laptop was probably your CIO or your CISO.

" That's not the case anymore. The minute you give Cloud Code or Cursor or anything local file system access, all of a sudden you've exposed all of those clear text credentials into agent context. Really great example of this, one of my colleagues just working with somebody who was setting up an agent to go read Gmail, and that agent had basically went and tried to authenticate to Gmail and got access.

" And they realized that the agent had actually dug down into two years of download history in their downloads folder, found the recovery codes, and used those to authenticate themselves to that system. It's just a great example. People don't think about this stuff.

Think about how much just legacy stuff exists in random places in your file system. The minute you expose that to an agent context, it can be indexed so quickly and be used for pretty much anything. And the thing is, again, it has a reasoning capability.

This isn't a traditional machine workload. You don't know how it's being used or where it's being exfiltrated to or any of that. And so one of the things that we're going to be really pushing on at 1Password is how do we secure those clear text credentials on your machine?

Today, we have an environments capability that allows us to secure clear text environments files, which are just a treasure trove of API keys. env file that you as a developer have on a laptop today. There could be a dozen different static API keys, all sorts of credentials in there to get access to different services.

That's just a fan-out point. The minute an agent can read that, they can get access to all of those services. Now multiply that by the number of different repos and projects you probably have on your laptop at any given point in time.

We're talking hundreds of clear text credentials on any developer workstation. That's a tremendous opportunity for uplifting security and one that we're super focused on, have been really focused on, and we're really excited about taking that a step further with some of the new capabilities we're launching this year. I love it.

Good stuff, Jeff. Excellent. Hey, we're almost out of time here.

Any other recent news or anything you want to share with the audience? I'm just really excited to welcome the Epona team to 1Password. We announced this acquisition last week.

Mm-hmm. We're really excited about where we're going with this opportunity and what they're bringing to the portfolio for us. So just want to give them a warm welcome, and thanks for having me on today.

My pleasure. Hey, Jeff Malek, VP GM of Developer and AI at 1Password. Look, like everything else with AI, it's forcing us to relook, rethink, re-engineer how we did things that we thought we had at least partially solved.

But I'm sure we'll be coming back to it again and again. Jeff, keep up the great work. Come back and keep us posted, okay?

Hey, thanks a ton, Alan. Have a great day. My pleasure.

Jeff Malek here on Techstrong TV. We're going to take a break. We'll be back in just a moment.