As enterprises shift from AI experimentation to full production deployment, the attack surface has fundamentally changed — and the security industry is scrambling to catch up. F5, an application and API security company, will be showcasing a pointed message at RSA 2026: protecting AI systems requires moving beyond model-level controls and into the infrastructure and API layers where risk actually accumulates. In this VMblog Q&A, F5’s VP of AI, Jimmy White, says the defining security challenges of 2026 are no longer about which AI model an organization chooses, but about how that AI is governed, routed, and protected at runtime across hybrid and multicloud environments.
At RSAC booth #6365 in the North Expo, F5 will bring that argument to life through live demos and a featured session from Field CISO Chuck Herrin on Wednesday, March 25. The presentation, “Real World AI Security: Protecting Enterprise Models, APIs, and Data-in-Motion,” tackles one of the most pressing emerging threat vectors: agentic AI. As autonomous systems move beyond generating content to actively calling APIs, moving data, and initiating workflows, White contends that token lifecycle management and least-privilege access aren’t optional enhancements — they’re foundational requirements for any organization serious about deploying AI safely at scale.
++
VMblog: Give VMblog readers a quick overview of your company and its core mission.
Jimmy White: F5 is an application and API security company focused on protecting and delivering the world’s most critical applications and AI workloads. As enterprises modernize their infrastructure and deploy AI across hybrid and multicloud environments, our mission is to secure every app, every API, and now every AI model and agent — without forcing organizations to compromise performance or innovation.
We believe security should operate at the same layers where applications and AI actually run: the infrastructure and API layers. That’s where risk accumulates, and where control must live.
VMblog: Where can attendees find you at RSA 2026? What’s your booth number, and what kind of experience can visitors expect when they stop by?
White: We’ll be at booth #6365 in the North Expo. When visitors stop by, they can expect several live demos showcasing how F5 secures AI models, APIs, and application traffic across hybrid environments. We’ll also be hosting a featured session, “Real World AI Security: Protecting Enterprise Models, APIs, and Data-in-Motion,” presented by our Field CISO, Chuck Herrin, on Wednesday, March 25 at 11:30 a.m. PT in Briefing Center North. The session will explore how AI applications expand the attack surface through APIs, models, and agents — and how layered protections including WAAP, AI guardrails for token-based traffic, and data-in-motion security help organizations protect production AI systems and maintain compliance.
VMblog: What were your key learnings from 2025’s security landscape, and how have those insights shaped your solutions for 2026?
White: 2025 made one thing clear: AI is no longer experimental. It’s operational. Enterprises moved from pilots to production AI systems — and that dramatically expanded the attack surface.
We saw three major shifts. First, AI systems are increasingly composed of multiple models, APIs, and external data sources. Second, token management became central to security, because every AI interaction is mediated by credentials and API calls. Third, runtime risk matters more than static model evaluation. A model that looks safe in testing can behave very differently under sustained, adversarial interaction.
For 2026, that means focusing on runtime protection, token governance, and securing AI at the infrastructure layer, not just the model layer.
VMblog: How is your company addressing the intersection of generative AI and cybersecurity – both defending against AI-powered attacks and leveraging AI agents for security operations?
White: Generative AI is accelerating both sides of the equation. Attackers are using AI to scale reconnaissance, automate phishing, and probe APIs at unprecedented speed. At the same time, defenders are deploying AI agents to automate response and reduce operational fatigue.
Our view is that AI security cannot rely solely on model-level controls. The real exposure point is often the API layer. That’s where prompts enter, tokens are issued, data is accessed, and actions are executed.
We focus on protecting AI systems at runtime: inspecting traffic, enforcing policy, managing tokens, and ensuring that AI agents operate within tightly scoped, auditable boundaries. AI agents must be automation-first, with actions evaluated, logged, and governed.
VMblog: With the proliferation of AI agents and autonomous systems in enterprise environments, how is your solution addressing the security challenges of agentic AI?
White: Agentic AI changes the security model entirely. These systems don’t just generate content; they take actions. They call APIs, move data, initiate workflows, and interact with other systems autonomously.
The biggest risk isn’t just prompt injection. It’s what happens after an agent is manipulated — what data it can access, what tokens it can use, and what systems it can touch.
We believe agent security must start with token lifecycle management and runtime enforcement. Agents should operate with least privilege access, tightly bounded scopes, and continuous evaluation of their actions. Without that, organizations risk creating highly automated systems that can unintentionally amplify mistakes or adversarial manipulation.
VMblog: What specific market challenges or pain points is your company addressing at RSA 2026? How have these evolved from last year?
White: Last year, most conversations were about “Which AI model should we use?” This year, the conversation is about operational control.
Enterprises are running multiple models across different environments for cost, performance, and compliance reasons. The challenge isn’t model selection. It’s model routing, governance, and secure execution at scale.
We’re addressing the need for secure AI data movement, resilient API layers, token management, and runtime observability. Organizations are realizing that AI infrastructure is becoming mission-critical infrastructure. It must be secure by design, not bolted on afterward.
VMblog: What sets your solution apart in today’s crowded cybersecurity marketplace? Why should RSA attendees prioritize visiting your booth?
White: Many vendors focus exclusively on model testing or AI red teaming. Those are important, but they don’t solve the full problem.
What differentiates F5 is that we operate at the infrastructure and API layer — where AI actually interacts with the real world. We secure the flow of data, the issuance and validation of tokens, and the runtime execution of AI-driven transactions across hybrid environments.
If you’re deploying AI in production, your biggest risks are not theoretical model flaws, they’re operational exposure across APIs and distributed infrastructure. That’s where we bring visibility and control.
VMblog: How does your solution help organizations address the expanding regulatory landscape, including AI governance requirements and emerging global privacy regulations?
White: Regulators are increasingly focused on accountability, auditability, and data governance in AI systems. That requires visibility into who accessed what data, which model was invoked, and what actions were taken as a result.
By enforcing token governance, runtime inspection, and detailed logging, we help organizations demonstrate control over AI workflows. Governance cannot be achieved through policy documents alone — it must be enforced in the infrastructure.
As AI becomes embedded in critical workflows, compliance will depend on operational controls, not just model documentation.
VMblog: What’s your perspective on the most critical cybersecurity trends that will shape the industry in 2026-2027?
White: Three trends stand out.
First, AI-driven automation will dramatically increase the velocity of both attacks and defense. Speed will become the defining variable in security outcomes.
Second, token management will become the primary control plane for AI systems. Every AI interaction is authenticated, authorized, and mediated through tokens. That makes token security foundational.
Third, we’re entering an era where emerging technologies — including quantum computing — will force organizations to rethink long-term cryptographic resilience. While quantum threats are still developing, forward-looking enterprises are already planning for post-quantum cryptography transitions.
The broader theme is this: security is shifting from perimeter-based protection to continuous, runtime-driven control. AI simply accelerates that shift.
##
