Okta Oktane 2026 | Nancy Wang, 1Password

Clips
More from Okta Oktane 2026

Nancy Wang

CTO

1Password

play_circle_outline Agent identity and authority: distinguishing human, machine, and agent actions

play_circle_outline AI-Driven Task Authorization: Mapping Agent Intent, Verifying Behavior, and Reducing Security Theater

play_circle_outline 1Password–Okta integration: protecting credentials and enabling secure SSO/device trust

play_circle_outline Credential Brokerage and Sandboxed Agent Execution

Info
Transcript

Nancy Wang, 1Password

Nancy Wang

CTO 1Password

In this interview from Oktane 2026, Nancy Wang, chief technology officer of 1Password, joins theCUBE's Rebecca Knight and theCUBE Research's Krista Case to discuss how Okta is extending identity security to AI agents as they become active participants in enterprise workflows. Wang explains why agents blur the line between human and nonhuman identity, acting on behalf of people or companies while carrying their own authority. She details 1Password's task-based authorization model, which grants just-in-time access and requires agents to prove they completed pri... Read more

explore Keep Exploring

Are software agents human or non-human identities, and how should their actions (for example in audit logs) be represented when they act on behalf of individuals or organizations? add

How do you ensure secure, task-based authorization and effective (scalable) human oversight for automated agents that perform many tasks at machine speed? add

How do Okta and 1Password work together for shared customers, what critical information needs to flow between them (especially for AI agents not bound by traditional SSO), and can you give tangible examples of how this improves access to sensitive systems? add

What role do you see credential brokerage playing in this industry, and might it become an architectural layer that sits between agents (or sandboxes/personal agents) and enterprise systems? add

bolt Powered by CUBE AI

Nancy Wang, 1Password

search

Rebecca Knight

>> Hello everyone and welcome back to theCUBE's coverage of Okta Oktane 2026. I'm your host, Rebecca Knight, alongside Krista Case, the principal analyst at theCUBE Research. Krista, one of the things that keeps me up at night is this idea of agents making very consequential decisions, either for me as an employee, as a worker, Or also in my personal consumer life, taking these decisions for something that's a legitimate task that it should be doing, and yet it does something I didn't necessarily want it to do.

Krista Case

>> And I think, Rebecca, we see this in terms of when we're asking them to take an action for us. We also see sometimes I ask it even just to write something for me, and all of a sudden it gives me back something I wasn't expecting. And so definitely it's cause for concern for sure.

Rebecca Knight

>> Yeah, so let's talk more about that with our next guest. She is Nancy Wang. Chief Technology Officer at 1Password. Thanks so much for coming back on the show, Nancy.

Nancy Wang

>> Yeah, thanks so much, Rebecca and Krista, for having me.

Rebecca Knight

>> So 1Password built its name protecting passwords for people. When did agent identity first show up as a problem that you had to actually solve? what was the raison d'être of 1Password?

Nancy Wang

>> Yeah, for sure. So as the holder and frankly the most trusted vault on the internet, we have been storing human passwords for over 20 years now. And of course, that has earned us the privilege to then also store developer secrets, machine secrets. So think API keys, right? SSH keys if you're a developer and you're wanting to push something to GitHub. And now, of course, agents, right? The topic du jour. And agents are really this interesting non-human identity. However, they can also appear as human identities because let's take the example of personal agents that has permeated the Twitterverse, right? Call it, well, they act as you. So when it goes and does something in a call log or an audit log, it looks like Rebecca did something, right? When in fact it was your agent. Or for example, the agent could also belong to a company. So for example, we've built our own incident response agent and that agent acts on behalf of the 1Password engineering team when it does incident response. And so you really have these two very different worlds. And so is it a machine? Is it a human? the right answer is it's probably both. Which is why it makes it actually even more important to understand what is the identity this agent is acting as. And even more so, what is the authority that it carries alongside the identity, right? When it goes about and does things. And this goes into sort of our true belief, which is whether it's for human users, machine users, or agent users or agents, for example, We want to make sure that we're not giving anybody standing privilege, right? And it's access granted just in time. And we also validate, right? Why do you need this access in the first place? Is it valid or is it not very valid, right? And so this goes back into how do we then trust that our agents are actually doing the right thing?

Krista Case

>> Yeah. And Nancy, so building on that concept, because I know obviously this is kind of a core foundational concept, at 1Password. When you think about the agent's authority, how do you think about making decisions dynamically about what they're allowed to do versus what they're not allowed to do in the event that maybe an agent starts to drift in some of their behaviors or maybe starts to consider taking some actions that we weren't anticipating?

Nancy Wang

>> Yeah, so that goes back into a design principle that we have at 1Password, which is task-based authorization, which means that, there could be workflows with many, many different tasks that could span over the span of many days. But that said, just like you would verify whatever an intern does before you allow them to move on to the next project, same thing with agents, right? You just want that agent to prove that, hey, it actually finished its last task with the right level of permissions, doing the right thing before you allow it to move on to the next task.

Rebecca Knight

>> So that's just making sure that there is in fact some human oversight in this that is watching what the agent is doing. But how do you make sure that the human is in fact paying attention to the degree you need that human to be paying attention? Because I think that that's what the concern is of all of this, right?

Nancy Wang

>> Of course. Similar with coding agents producing so many PRs, if a human were to sit on the other end and read all of those, all of that code, right? There'd never be enough hours in the day for that. And so this is where we have some really exciting launches coming up in the next couple months with some of the model labs on leveraging applied AI to be able to say, hey, this is the intent in which the agent is acting and can we map it back to the role, right? Can the identity of the agent and the authority that the agent is carrying with it as it goes about these tasks? And is that legitimate and valid, right? And we can certainly when it launches, go into the architecture that we built to provide confidence for our human users because to your question earlier, Krista, that is the number one request from our human users because agents acting at machine speed, they're probably going to make thousands of calls every minute. Right. And if you have a human on the other end saying approve, approve, approve, well, that just becomes security theater after some point because you just can't possibly keep up. Yeah.

Krista Case

>> And Nancy, I'm glad that you brought that up because I think this concept of agent swarms is something that's particularly potentially concerning in this conversation. And so when you think about delegating authority, From your perspective, what needs to happen if an agent is trying to delegate authority to another agent? How do we think about putting guardrails around that?

Nancy Wang

>> Yeah, so it goes back to what is a swarm spun up to do?

Krista Case

>> Yeah.

Nancy Wang

>> And so we've done various experiments and now moving into production at 1Password, and one of the swarms that we created internally was to, for example, debug a live production database, make config changes, right? And so how do we do that? Well, each agent, for example, has its own identifier within that swarm, right? But overall, the authority of that swarm is to act on behalf of 1Password engineering as it makes various requests. But that said, right, if you want to isolate the specific action to a particular agent within that swarm to say, hey, you made the wrong config change, right, or you looked at the wrong database, for example, you need the ability to go back in time and essentially do that. And so that is why being able to not just have all of the agents in a swarm share the same identity, but actually have unique identifiers that is bound cryptographically to the intent such that as the agent traverses from, let's say, data plane to data plane or control plane making various task-based requests, right? You can track essentially that, we call it chain of trust throughout the agent's actions in that workflow.

Rebecca Knight

>> Tell us a little bit about the Okta, 1Password relationship and how you would change something in the life of a shared customer in a way that's really tangible.

Nancy Wang

>> Absolutely. Well, look, I believe Todd and our founders have known each other for some time. The identity world is very small. And one of the amazing things is that we have tons of joint customers who have both 1Password and Okta, right? Because as we know, SSO is pervasive. As one of the securest ways to log into applications. Now that said, there are also a whole slew of applications that aren't behind SSO that require credentials. And especially as we're talking about agentic development that more and more so contains, for example, tokens or keys, right? And so we wanna make sure that whatever method our users are gaining access to sensitive systems or applications, that they are protected along the way. And that's really the joint relationship between Okta and 1Password.

Krista Case

>> Yeah, and so I think when you think about the information, Nancy, that needs to kind of travel between 1Password and Okta, can you talk about what are some of the critical points of information, especially for an AI agent that to your point is not necessarily bound by traditional SSO?

Nancy Wang

>> Yeah, absolutely. So today, for example, we have very complementary products in the sense that oftentimes, the Okta login is stored within 1Password. And so today when a user, for example, wants to log into Okta to get access to their federated applications, they can do so using 1Password. At the same time, for example, we also have joint customers who use 1Password Device Trust as a multifactor to get into Okta. So essentially you could say, hey, only these users with these types of managed devices or devices that are very healthy, because we can see the health posture of your devices, can get into Okta and therefore some of the most sensitive and critical applications within your enterprise.

Krista Case

>> And Nancy, when you think about the Blueprint Alliance and how you're working with 1Password, we've been having a number of conversations about the Blueprint Alliance here in theCUBE today. And we're talking about, you know, there's the interoperability piece of it, but there's also the need to be able to agree on what that trust, that intent, and that delegated authority means at the end of the day. So can you talk about maybe how the two platforms are rationalizing with each other to make sure that there isn't a conflicting decision about that?

Nancy Wang

>> Of course. This is where standards for the identity industry are so important, right? And we are proud stakeholders, board seat holders of the FIDO Alliance as an example. But certainly this is where all the work that Okta has done around XAA or ID-JAG comes into play. In fact, we were just having a conversation with Karl McGuinness around how he's thinking about making sure that these protocols become industry standard. And certainly we want to make sure that wherever, as our customers, for example, joint customers with Okta, are authenticating in, that they are also able to access 1Password, right? And also as the agent traverses across the ecosystem, that that chain of trust can be shared between the two companies.

Rebecca Knight

>> Our conversation is mostly centered around workflows, but there's a big part of this that's getting into more of the consumer life. And I know that your team in particular is doing some work there. Can you talk a little bit about how you view these challenges from the consumer perspective and if you view them differently? And also in your personal life, what you're empowering agents to do on your behalf?

Nancy Wang

>> For sure. I feel like you certainly have some thoughts as well, Rebecca. So I'd love to hear your thoughts as well. Well, look, I think with every wave, right, agents— first wave was really seeing coding agents then came knowledge worker agents. And now, of course, it's the age of consumer AI. And so with consumer AI, there's so many workflows that you and I do, right, every day as consumers, as parents, right, as friends, family calling the doctor, making sure that you book that table at that restaurant you really wanted, right? Or you fill out this visa form because you're traveling to a country. A lot of that is tedious and very manual, right? And requires many different prerequisite steps that sometimes it's just impossible to serialize despite your best intentions. Well, guess what? A personal agent can serialize that for you such that when you're saying, hey, let me go fill out this visa form, it actually has everything that you need at your fingertips. Maybe that picture resized, right? Or the specific addresses that you need or the location, the hotel.

Krista Case

>> Right, but there's also different degrees of risk inherent in even just these use cases that you're describing, Nancy. So for example, if I'm going to book a dinner reservation, I need to give it my name and my phone number. That's potentially a very different risk than if I am using it to maybe fill out a visa form, for example. So maybe how do you think about that in terms of distinguishing when you're okay with letting a personal agent go ahead and run with some of your information?

Nancy Wang

>> Of course, and similar with credentials, right? If you don't give an agent your credentials, well, guess what? It's just a glorified chatbot, right? So the only way it can be productive is if it gets access to the things that you as a human normally would have access to. Well, but that also goes into where our users are now pulling us into the conversation of, hey, we want to use this amazing technology, right, to automate my life, to go through the manual, tedious workflows of coordinating my kids' after-school activities, for example, right, for busy moms. And we want to be there to help them, but do that in a safe way. And this is where, we are actively working with, and you've seen some of these pre-announcements from, Cursor, from Meta Muse, from Instinct, and there'll be many, many others coming in the next couple of weeks here where it's actually using 1Password under the hood, which means that it's our same guarantee that we launched when we integrated with Claude Code, right? Which is we never show the credentials to the agent or the underlying model. And also we have to authorize each agent request to credentials that are in 1Password, right? And so that could be a secure way for our users to leverage the power of these personal agents, but not give it, for example, sensitive things like your Social Security number or your credit card information.

Krista Case

>> And that— I'm glad you brought that up, Nancy. I actually had a conversation with 1Password a couple weeks— a couple months ago, actually, at this point, back at Black Hat. And we were talking about credential brokerage and the need to, as you mentioned, reduce the possibility of exposing those credentials to the model. So what is the role that you see credential brokerage as having in this industry? Do you think it's going to become kind of an architectural layer that sits between agents and enterprise systems?

Nancy Wang

>> Absolutely. In fact, there are also conversations that we're having with some of the leading sandbox providers as well, because if you think about, frankly, a sandbox or a personal agent as simply an ecosystem, where we can— or a client where we can safely provide access to sensitive systems for either agents or for agents running in sandboxes without, again, revealing the credentials inside that sandbox or to that agent. It becomes a secure way for agents to go do things that require credentials, but without actually giving the agents exactly in the prompt, hey, this is my API key, for example.

Krista Case

>> Yeah.

Rebecca Knight

>> So what do you think we're going to be talking about at next year's Okta Oktane, because there are so many, as you say, new product announcements that are going to be coming out in the next few weeks on the consumer side. And of course, organizations are moving so quickly to deploy these kinds of technologies and have agents taking ever more consequential decisions on an organization's behalf and a consumer's behalf. So what are some of the things that you're talking about here with your fellow partners and fellow customers and the things that you're going to take back with you on Monday when you get back to work?

Nancy Wang

>> Well, look, we work very, very closely with all of the model labs, as you've seen in our announcements. And so certainly this is probably a cheat code because I do have a peek into the future and what's coming up is, look, I think that all of our apps are going to become thin clients and that most of the code will be written in remote sandboxes. And so how do we prepare for a world where, things are happening in the cloud? Right. Secure access needs to happen in the cloud. And that's going to be an exciting future where truly we're going to be able to unlock the productivity increase of long-running workflows because they no longer need to be powered by your laptop.

Rebecca Knight

>> And organize our kids' activities in a much more efficient way.

Krista Case

>> Exactly.

Nancy Wang

>> While we're sleeping, while we're at the spa.

Krista Case

>> Right, right. Exactly.

Rebecca Knight

>> Right.

Nancy Wang

>> Exactly.

Rebecca Knight

>> Well, Nancy, thank you so much for coming on. A really interesting conversation.

Nancy Wang

>> Thanks so much for having me.

Krista Case

>> Thank you.

Rebecca Knight

>> I'm Rebecca Knight for Krista Case Stay tuned for more of theCUBE's coverage of Okta Oktane. You're watching theCUBE, the leader in enterprise tech news and analysis.