Security

RSAC 2026: OpenClaw and the Rise of Agentic AI Security

As AI agents proliferate, frameworks like OpenClaw and Cisco’s DefenseClaw signal a new cybersecurity paradigm built for autonomous systems operating at scale.

Joe Kuehne

Joe Kuehne is an Associate Editorial Director at BizTech magazine.

Nearly as soon as NVIDIA GTC came to a close, RSAC began. So, it’s probably no surprise that some of the major themes and announcements from the first event carried into conversations at the second. And one of those themes was the pressing need to secure agentic artificial intelligence.

The cybersecurity landscape is undergoing a profound transformation as organizations enter what Cisco president and chief product officer Jeetu Patel calls “one of the most exciting and simultaneously the most disorienting times in human history.”

At the center of this shift is the rise of agentic AI — autonomous software entities capable of planning, reasoning and executing tasks with minimal human intervention. While these agents unlock unprecedented productivity, they also introduce entirely new categories of risk.

“Agents plan, they act entirely autonomously,” Patel explained at RSAC 2026. “We shouldn’t think of these agents as tools. We should think of these agents more like digital coworkers.”

This evolution demands a rethinking of cybersecurity itself. Traditional models built to protect human users and static systems are insufficient for a world where millions of agents interact dynamically across systems, application programming interfaces and data environments.

Click the banner below to lay the data governance foundation needed for AI.

Why Agentic AI Requires a New Security Model

Unlike chatbots, which primarily generate outputs, AI agents take action. That distinction dramatically increases risk.

“With an agent, you have to worry about it taking the wrong action,” Patel noted, and taking an action that could potentially be irreversible.

Cisco’s own research underscores the urgency. According to a March 23 press release by the company, while 85% of enterprises are experimenting with AI agents, only 5% have deployed them into production, largely due to security concerns.

To address this gap, Cisco has introduced a new framework for securing the “agentic workforce,” built on three pillars:

Protecting the world from agents
Protecting agents from the world
Detecting and responding at machine speed

These principles form the foundation for emerging technologies like OpenClaw and Cisco’s newly introduced DefenseClaw.

WATCH: Check out the cybersecurity trends to watch in 2026.

OpenClaw, DefenseClaw and NVIDIA’s Open Ecosystem

At RSAC 2026, Cisco announced DefenseClaw, an open-source framework designed to secure AI agents throughout their lifecycle. The framework integrates tools such as skills scanners, model security checks and automated inventory systems to ensure that “every skill is scanned and sandboxed, every MCP server is verified, and every AI asset is automatically inventoried,” according to the company’s press release.

Crucially, DefenseClaw is designed to integrate with NVIDIA’s OpenShell — announced at NVIDIA GTC — as a secure runtime environment for agent execution.

This combination reflects the broader OpenClaw ecosystem:

OpenShell (NVIDIA), a secure containerized runtime for AI agents
DefenseClaw (Cisco), a security framework that embeds guardrails, validation and monitoring
OpenClaw approach, an open, interoperable model for deploying secure, scalable agent systems

Together, these technologies aim to eliminate manual security steps and enable organizations to deploy agents safely at scale.

The difference between delegation and trusted delegation is the difference between bankruptcy and market leadership.”

Jeetu Patel President and Chief Product Officer, Cisco

Security at Machine Speed

One of the defining challenges of agentic AI is scale. Patel warned that organizations could soon manage trillions of agents, each interacting with sensitive systems and data. Human-driven security processes simply cannot keep up.

“If we were only limited to responding at human scale, we’d be on the cusp of an epic catastrophe,” he said.

This is where OpenClaw-style architectures become essential. By embedding security directly into agent workflows and runtime environments, these systems enable:

Continuous validation of agent behavior
Real-time policy enforcement
Automated detection and response at machine speed

Cisco’s integration of AI into security operations centers reflects this shift. Agent-driven SOC tools can triage threats, correlate data and execute responses automatically — transforming security from reactive to proactive.

Balancing the Push for Automated Security With a Bit of Caution

Nearly all of the cybersecurity leaders speaking at RSAC acknowledged the need for automation in an organization’s security strategy; human cyber defense teams just can’t work as fast as fast as AI adversaries. But some leaders are proceeding with caution.

Speaking at RSAC, Commvault Field CTO Vidya Shankaran emphasized the need to strike a balance: “Automation is a must-have tool in your toolkit, because it's humanly impossible if you're trying to reach that scale of your operations. But I wouldn't hyperautomate. Hyperautomation definitely produces a lot of blind spots.”

George Kurtz, CEO and founder of CrowdStrike, also tempered his embrace of AI with a dose of prudence.

“We need the right guardrails. And if we leave this to chance, or we leave this to regulation, or we leave this to somebody else, it’s going to be problematic,” he said. “So, my call to action is for everyone to take command, to think about AI safety, to be able to be conversant to the board, and to be able to go back to your businesses and say, ‘Hey, we are AI realists. We want to help you. We want to promote AI. We just need to do it in a safe and responsible way.’”

Click the banner below for deeper insight into modern cyber resilience.

Trust as the Defining Challenge

Ultimately, Cisco and Patel theorize that the success of agentic AI hinges on trust. “The difference between delegation and trusted delegation is the difference between bankruptcy and market leadership,” he emphasized.

OpenClaw and related frameworks aim to operationalize that trust by ensuring:

Agents have verified identities
Actions are authorized and constrained
Behavior is continuously monitored
Risks are dynamically managed

This aligns with Cisco’s broader vision of extending zero-trust principles to AI agents, treating them as first-class participants in enterprise environments with strict identity, access and governance controls.

Photography by Joe Kuehne