The recent launch of the F5 AI Security Platform aims to help partners and customers achieve improved visibility and security for AI usage, F5 CEO François Locoh-Donou tells CRN.
‘Enormous Market’ For AI Security
F5 is doubling down on enabling solution and service providers to capitalize on surging AI adoption through the recent launch of its unified platform for discovering, testing and securing AI models, according to F5 CEO François Locoh-Donou.
In an interview with CRN, Locoh-Donou, who is also chairman and president at F5, said it’s clear that “AI is quite a vulnerable technology today.”
[Related: F5 Advances Security For The AI Era And The Post‑Quantum One Coming Next]
“These AI models are still vulnerable, he said. “And yet they’re being deployed at scale—because enterprises cannot wait to start getting the benefits, the return on investment, the productivity gains of AI.”
In late June, the vendor announced the launch of the F5 AI Security Platform, in connection with the company’s acquisition of SurePath AI. The new platform brings together SurePath capabilities for boosting AI visibility with AI red teaming functionality and custom guardrails from its prior acquisition of CalypsoAI. The ultimate goal is to help partners and customers identify the AI models that are being used, before then testing them for vulnerabilities and addressing the risks, Locoh-Donou said.
“We’re detecting, for you, what AI models are being used by your employees or your business. Then we have the ability to do penetration testing on those AI models that are being used in your organization and find what vulnerabilities exist in them,” he said. “And then we have the ability, once we’ve done that, to remediate those vulnerabilities with our guardrails—to ensure that [for] all your models, you know what you’re using, you’ve tested them and you’ve put in place the security to protect them.”
Looking ahead, “we’re going to continue to invest in that,” Locoh-Donou said. “It’s going to be an enormous market.”
The growing opportunity also comes amid an accelerating shift among large enterprises toward building their own on-premises AI infrastructure, and even doing cloud repatriation in some cases, according to Locoh-Donou.
“I think we’re likely to see a cycle of infrastructure buildout where companies build their own AI infrastructure, their own GPUs, to run their models on GPUs themselves rather than paying public cloud providers to do that for them,” he said. “Not every company will do that. But I think some of the large companies will seriously consider, or do that, themselves. And that will be another source of repatriation.”
What follows is more of CRN’s interview with Locoh-Donou.
We’re about midway through the year—what are the major trends of the year that you would point to in your space?
One is, hybrid multi-cloud is becoming this de facto architecture. Several years ago, it used to be that everybody thought they were going to be all-in on the public cloud, maybe all-in on a single cloud. Today, 94 percent of enterprises are embracing hybrid multi-cloud architectures. That’s a significant tailwind for F5, because we have always been advocating hybrid multi-cloud and that customers needed the flexibility to place their applications in the most appropriate environment. That trend has accelerated internationally with digital sovereignty and a lot of companies wanting to have autonomy over their own stack, their operations, their data. And so some are repatriating from the public cloud—the big hyperscalers—onto either local clouds in the countries or on-premises data centers. And so hybrid multi-cloud is driving enterprise reinvestment in data centers. So that’s really a good trend for F5 that we have worked hard to position ourselves for.
Then, the second big trend we’re seeing is that the threat landscape is expanding. The amount of attacks—particularly on applications and APIs, which is where F5 is focused—are expanding. And, of course, AI itself is becoming an attack surface. All of that creates a need for more security solutions like the ones F5 is focused on, specifically as a security leader in app and API security—and, increasingly, in AI security. And then the third big trend I would say we have seen this year is, of course, AI infrastructure in general—the buildout of AI infrastructure and the implications that has for AI-enabled applications and hardening data pipelines for companies that are trying to feed all this data into their AI models. That’s also an area where F5 has a strong position.
So those three trends—hybrid multi-cloud, the expanding threat landscape and AI—I would say we started to see them a little last year, but I think they’ve been accelerating and have been clearer this year.
Is repatriation something that you’re seeing a lot of, or more than you would have expected given the environment?
Yes, I think we’re certainly seeing more of it than we were, say, three years ago. And if you survey enterprises, I think more of them have plans and expectations that they will repatriate more workloads in the future. There are a few things driving that. Internationally, I mentioned digital sovereignty is a factor. The geopolitics of the last three to five years have really sensitized a lot of countries to the need to not depend too much on single infrastructure providers, especially single infrastructure providers that are U.S.-based. And so they are repatriating to have their [resources] on-premises in their own data centers. And we’re seeing more data center buildouts as a result of that, or they’re choosing local clouds—country-based clouds—that are completely sovereign clouds to do that. Now, probably the big public-cloud hyperscalers will respond to that, by building “sovereign clouds” in multiple countries. Whether that response is going to be effective is [still] to be seen.
Beyond digital sovereignty, companies have also realized their data is more valuable—because, in the world of AI, data is the raw material. So they’re collecting more data about their business, their customers, their product, etc. A number of them who have stored this data in public clouds are realizing that collecting all that data and having it in the public cloud is expensive. The ingress and egress cost of moving data in and out of the cloud is very expensive. And it’s actually cheaper to have this data on-prem and close to the infrastructure pipelines that they already trust. So that’s another source of repatriation.
And I think one thing we are starting to see—it’s a nascent trend—is a number of companies may conclude that they are better off having their own AI infrastructure. At the moment, I would say the vast majority of companies that are doing AI are renting GPUs as a service, largely from public cloud providers, from the big hyperscalers. But that’s expensive. There are reasons to do that today because GPUs are evolving very quickly. And so if you have to replace this thing every 18 months and your depreciation cycle is five years, it doesn’t work. So it’s better to rent them from cloud providers, but it’s also more expensive. I think we’re likely to see a cycle of infrastructure buildout where companies build their own AI infrastructure, their own GPUs, to run their models on GPUs themselves rather than paying public cloud providers to do that for them. Not every company will do that. But I think some of the large companies will seriously consider, or do that, themselves. And that will be another source of repatriation.
What are the biggest things to know about your new AI security platform and how that is differentiated from other offerings?
Our focus over the last decade has been securing applications and APIs. We’re continuing to see substantial growth in app and API security. The strength that we have brought to that security market is that F5 has been infrastructure-agnostic—in the sense that we can secure any application or API, anywhere, whether it’s in the public cloud, whether it’s in an on-premises data center, whether it’s at the edge or colocation. We’ve built a platform that is agnostic to these infrastructure considerations. And we’re doing the same thing in AI security. The competencies that we have built for decades in app and API security are very applicable to AI security. We are basically the player in the market that is specialized in fine-grained security—understanding the semantics of traffic and the logic of applications. Ultimately, those capabilities—it’s called Layer 4 to Layer 7 security—are really critical in AI security because you eventually have to secure every token. So we brought very strong DNA of this fine-grained security to the table. Being able to do fine-grained security, in real time, is very difficult. And I think F5 really is a leader in that domain. Now, on AI security, my opinion on that is that AI is quite a vulnerable technology today. These AI models are still vulnerable. And yet they’re being deployed at scale—because enterprises cannot wait to start getting the benefits, the return on investment, the productivity gains of AI. And so we have assembled this AI security platform to help customers deploy AI models—yet at the same time, do so in a way that is responsible and secure.
What are the biggest components of the platform that you’d want partners and customers to know about?
You don’t want customers to have to go and buy point solutions here, point solutions there—which is a lot of what happened in security over the last 20 years. Customers ended up with a lot of complexity. And so that’s part of why you’ve seen us make some acquisitions to build this AI security platform. We acquired, a few months ago, Calypso, which brought us red teaming—penetration testing of models—and also the ability to remediate the vulnerabilities that we find with what’s called custom guardrails. Essentially, we today have the best custom guardrail solution in the market. But that’s not enough. You also need to help customers discover what AI models are being used in their organization, either by employees or by certain workflows. And that is why we bought SurePath AI. That brought us visibility into what AI is being used in an organization. And visibility obviously is really important because you cannot protect what you can’t see. And so F5’s AI security platform now has visibility. We’re detecting, for you, what AI models are being used by your employees or your business. Then we have the ability to do penetration testing on those AI models that are being used in your organization and find what vulnerabilities exist in them. And then we have the ability, once we’ve done that, to remediate those vulnerabilities with our guardrails—to ensure that [for] all your models, you know what you’re using, you’ve tested them and you’ve put in place the security to protect them. All of that is part of this F5 AI security platform. And we’re going to continue to invest in that because AI security, if you look at the predictions, it’s going to be a very large market. It’s already $45 billion today. I think it’s going to grow to $160 billion over the next three to five years. So it’s going to be an enormous market. Customers are going to want to protect AI—already, customers are starting to become very sensitive to that. So we see a substantial opportunity there.
What are the biggest opportunities for partners in working with F5 in 2026?
First, for our partners, is the work we have been doing in app and API security. That market continues to grow. We are seeing more attacks. Generally, the number of attacks is growing rapidly on enterprises—but also, the percentage of those attacks that are happening at the application layer and the API layer is growing. It’s concentrating more on that, which is where we and our partners have been focused. So I think the first imperative for our partners is to continue to drive those conversations with our customers. And those conversations naturally lead to the next application, which is an AI model, and the next API, which is an agent API. How do we secure that?
We’re working with our partners to enable them to be able to have these conversations. We’ve done technical training already. We did an AI summit in June. We’ve launched the technical foundation program for our partners to get them up to speed and retrain them on our product families. We’re launching another AI-specific enablement program for partners later this summer. So there’s a lot of work we’re doing with partners to make sure that the right knowledge is there for them.
What are the biggest differentiators you would point to, in terms of partners working with F5 in this environment?
The differentiation is, No. 1, we are putting together a complete platform. No. 2, as we have done in the past for other things, it is infrastructure-agnostic. If a customer wants to run it in the cloud, we can do that. If a customer wants to run it as-a-service in the cloud, we can offer it to them as-a-service. If they want the software or hardware to be deployed in their environment, on-premises, we can do that. Not every vendor can do that. Most vendors can do one or the other. But F5 has always had this philosophy of, we’re going to be agnostic to the infrastructure and where you run your agents or where you run your models. Because ultimately, AI is very hybrid—so customers are going to want to run AI models on-prem and in the cloud. Agents are going to be on-prem and in the cloud. And we can’t put constraints on customers as to where they run their AI. We have to adapt and give them the flexibility with an AI security platform that supports all environments. That’s differentiated—that’s specific to F5.
Our protection of AI models is the most advanced in the market. We have agents that simulate thousands of attack patterns every month on AI models, including the most obscure AI patterns. We add 10,000 attacks to our library every single month with our agents. And we have the ability to build these guardrails that are custom-made for a given AI model.
Most other vendors have this one-size-fits-all set of guardrails. But F5 has a technology that basically allows us to customize these guardrails. They’re programmable and custom-built for a customer’s AI model. And that is a very powerful capability because all these AI models are different, and they can be attacked differently. A model about life sciences is not going to be attacked in the same way as a model about banking or credit decisioning. And so being able to customize those guardrails to the particular thing that a model does is a powerful capability of F5.
