I can highlight a few disruptions that are in play in the market right now. Agentic identity or agentic AI is absolutely one of those, along with the risk of deepfakes.

Quantum-safe computing is a huge issue. I think folks are starting to see the horizon where quantum will become real, and they are starting to implement quantum-safe crypto systems to protect data.

The final emerging technology is what I’ll call wallet-based trust. If you think about how consumers today are interacting digitally with the world, particularly with tap-and-go payments for boarding airplanes, driving cars, and getting access to concerts, all kinds of things are happening via their smartphone wallet. These credentials in their wallet are are for verification. For example, that they are old enough to buy a beverage or that they are of a certain age to operate machinery. The use cases are endless in terms of how these credentials need to be verified and are going to be utilized. Importantly, the consumer expectation for the experience has reached a tipping point where it’s almost becoming the de facto way that a consumer would like to interact with the world. It’s like when we moved from cash to credit cards in the checkout lines. It’s now very frustrating to see somebody pay with cash, rather than just tapping and going. I expect the same thing to happen with identity, wallets, and credentials, where the current username and password experience is going to change. The thing they are going to log into is their wallet, providing a very seamless experience for them.

Peter Barker, Chief Product Officer, Ping Identity

The basic principle is that you can’t secure what you don’t know. If someone is using an AI application, you want to know its use is secure. You don’t want to be the last one that knows about it. The preference is to know about it in advance.

Putting things into the context of a modern security team—in just one week, everything can change. New applications running architecture changes. Visibility is about actually knowing what is running right now without the need to depend on a developer showing you a slide deck. Even if they remember to come to you and show you the deck, who says the slide deck is actually true? And even if it’s true, the situation might already have changed.

We can’t continue to rely on classic approaches where everything is static. Visibility means seeing across your cybersecurity environment in a way that is dynamic and that actually shows you the real attack surface.

Ami Luttwak, Co-founder and CTO, Wiz

From a detection process, I think incident response will stay as is. Ultimately, if something really blows up, if the top-shelf attacker shows up, humans will be doing the work. I don’t see that being transformed all that much.

But in the other parts of detection, I do see a lot of AI brainpower being applied. I think that large organization leaders may have a part in many of the tasks dealing with clearing the alerts—contextualizing them, understanding them, and then deciding what to do.

How it will be done is to be determined. It may still involve humans, but that part is where many of the startups are investing. They’re trying to apply superior brainpower and superior context knowledge of various labs to understand this. And the second bucket is turning vague intelligence into robust rules. I think many people have a dream of using AI to look up some PDF online that says APT42 can gain access to your network, and then a beautiful SIM-ready detection rule would come out. And it would detect that threat and would not produce false positives. If you told me this story two years ago, I would say that it’s a really good dream. Today, I pause before I give an answer because I’ve seen enough people that have converted loose, vague signals into threat intelligence.

Dr. Anton Chuvakin, Security Advisor at Office of the CISO, Google Cloud Security

What if these new methods, with this type of quantum computing power, were able to attack an organization in five days instead of five months? I think that the element of this is lost on a lot of people who are still arms crossed and quite skeptical. I would say that skepticism is slowly dissipating.

We have to get ready for this. The question is, how much effort do we need and what resources do we need now so that we are ready for this? We’re not sure when quantum day is. Is it when everybody now knows that there’s a cryptographically relevant quantum computer (CRQC) or is it that we’ve had a few hacks?

I don’t think you can afford to sit around anymore. Whether it’s a semblance of a discovery starting off or a full MRI of your organization, you’ve got to do something.

Blair Canavan, Director, Alliances – PQC Portfolio, Thales

A lot of people are confused about AI because they think the future of AI is that human work will be completely done by AI. AI assisting humans is a misnomer. For it to really work, humans also have to do work.

AI is taking all the world’s knowledge and everything that has ever been written or thought about, and it can correlate things and give you that knowledge in a succinct answer. But it can’t connect the dots, which humans still need to do.

In a world where a human is doing intuition work and the AI is doing the knowledge work, you have to ensure that the AI is secure, because if the knowledge is wrong and hacked, then intuition will be totally false. In that world, I do not believe this mode is going to work.

My vision of the cyber future is truly inside out. Security is a truly autonomous, disconnected, loosely coupled system, but every entity in the loosely coupled system is self-secure. Zero trust goes in that direction. Think about this autonomous AI agent that is your knowledge worker. If it is secure and it can only talk to the fully authenticated, trusted entity, then you have really solved the cybersecurity problem.

Then all these autonomous things are independently secure. Your orchestration of this autonomous agent is going to be secure by definition. Everything will assume that everything else is compromised. How do I talk in a trusted way to a trusted, authenticated entity? That’s going to be the future of AI.

Bipul Sinha, CEO, Rubrik

I think there are three categories that we believe are important from a minimum viability and a cyber resilience perspective. The first actually has very little to do with recovery, which is there are things that operationally organizations can do that are proactive and would help them recover faster but have really little to do with the actual recovery. For example, if a set of system patches is out of date. Or there were other policies that hadn’t been applied to a fleet or part of an estate, or if there’s been drift in a set of systems, these are all things that are going to cause issues upon recovery. There are also things smart agents can flag, such as, “Hang on a second, you’re running out of policy here or this is going to cause problems.”

Proactively finding those sorts of things—I think that’s a category where we believe a lot of focus should be applied. Another one I think of is around the mechanics of recovery, especially around the idea of clean point validation. What are you doing to scan for threats and anomalies and correlate across systems?

The last is that there are a lot of processes in the recovery space that are still manual: being able to use smart agents to go and clean your room, figuring out what your runbooks should be, and coding those things for you to help automate some of those processes.

Tim Zonca, VP of Portfolio Marketing, Commvault

We have capabilities that we’re enabling from an AI perspective where it’s going to give you recommendations. It sees trends on different types of playbooks. Let’s say Deloitte creates 10 to 20 playbooks for a customer. The AI agent can help an analyst find the best playbook for a particular situation.

It’s going to be able to call that out to the analyst. Even if it sees misconfiguration or is seeing something not right starting to bubble up, the analyst can now talk to the assistant and say, “Show me the top 10 threat actors that we’ve seen.”

It’s going to bubble that up from a language perspective. You can even bring your own machine learning (ML) under the covers. If you do data classification a certain way that only your organization does it, the agent could align that under the covers with the ML functionality, creating an ultimate easy button.

Anthony Polzine, Senior Global Solutions Architect, Palo Alto Networks

I think we are really leaning into the idea of AI agents as game changers. Whether that’s an agentic workforce or large language models (LLMs). What we’ve seen at scale with many of our customers is applications getting deployed all of a sudden, with questions then coming up from compliance and from security. Should these people really have access to this data? What controls do you have in place? How are you sure that you know? Can you show me the effective access?

We’ve seen this explosion of enthusiasm around using AI to make businesses more efficient. But then seeing many programs come to a screeching halt where they realize “I just can’t control this. I’ve got to rethink how this works.”

How do we get in front of that problem? How do we effectively show the access that someone has? If you are using AI agents, how do you have controls in place around the data in applications? You can have conditional access to certain people, giving Sarah a different view than Bob.

Access that’s appropriate to their role to help the business use agents to get the value they were hoping for and with control from the beginning—as opposed to having to come in two years from now and retrofit all those controls in place. I think more out-of-the-box solutions that help understand the landscape of agents are definitely on the horizon for the next few years.

Chris Gossett, Chief Growth Officer, SailPoint

I’ll answer that in two ways. Today, I would suggest most cybersecurity teams should be able to answer these questions about AI.

Do I have a complete inventory of all AI agents operating in my environment, including third parties? Do I have a process in place today to validate and evaluate the AI agent before it goes live into my environment? Do I understand the identity and access management, and authorization of those agents, to be able to communicate with other agents in my environment? How am I monitoring the health of those agents? Are they introducing new risks into the environment?

The biggest concern is, OK, so what happens when you’re done with your agent? Are we back to the cloud world where we have all these tenants that are dormant out there? The difference now is these dormant agents can be weaponized very easily.

Another example. You might decide to look at using a travel agent AI. Why does anyone need to use a travel agency anymore? You’re just out there looking for the best prices. Gather up the information and book a flight. You can automate the payment and everything else. Well, someone can also introduce fake information into that ecosystem. And suddenly you’re paying for things that actually aren’t real. Understanding how AI works, you can work to prevent introducing things that can cause it to capture false information and come to false conclusions.

Rich Baich, CISO, AT&T

The short answer is absolutely. The way I hear our clients talk about this is that everyone went through an exploratory phase in so many different ways. Now everyone seems to be shifting into how they narrow that focus.

They have gone from exploring everything to finding the high-value ROI use cases that they’re able to really focus on. Then they take a couple of those and really get live and in production. In general, I see that shift happening. I think there’s a couple of substantial technology shifts that have happened over the past 12 months that are enabling that, in a more material way as well.

If you think about 12 months ago, when everyone talked about AI risk and that was top of mind, most people talked about the models. Concerns about what is the bias in the model, hallucination risk, and quality. We’ve sort of moved past that. Those are still real material problems we have to solve and worry about, but it’s not as top of mind now with new AI updates.

It’s very easy to connect to your enterprise systems and data. That has sort of shifted to this new problem. Mobile AI needs data and data needs governance. You’ve got to solve for the security of it, and you’ve got to solve data governance of it.

Blake Brannon, Chief Strategy Officer, OneTrust

When thinking about why organizations have AI security guardrails in place, it’s a good analogy to think about why cars have brakes. Cars have brakes so that you can drive faster. When you think about AI defense guardrails, as an employee of an organization, if you know that your access to applications is protected with guardrails, you can do your job faster without thinking twice about whether you should really be using it or if you’re violating a policy.

Same with cloud applications. If you talk to most employees within large organizations, they want to do the right thing. They want to build better applications for their customers. The challenges around security are that a lot of the ways in which a model can be compromised are very difficult to understand, even for a technologist. If every time you’re working on a model, you have to think about the ways in which it might impact your organization, it can slow down your development life cycle. That’s where having a product actually enables faster velocity. What we hear constantly from VP of AI, VP of engineering, and VP of infosec is that it’s taking some of this weight off their shoulders.

Anand Raghavan, VP Products, AI, Cisco