Cisco expands AI Defense and AI-aware SASE for agentic enterprise security

Cisco said on February 10, 2026 that it is expanding AI Defense and rolling out AI-aware SASE capabilities as part of a broader security push for what it calls the agentic era. The core message is clear: enterprises are no longer only securing users, devices, and cloud apps. They now also have to secure models, prompts, agents, and the growing number of AI services that employees and internal teams can reach from across the network.

The company said AI Defense will extend across model development, deployment, and use. That matters because security risks show up at each stage. During development, organizations need visibility into the models and components they are adopting. During deployment, they need guardrails around policies and integration paths. During actual use, they need ways to monitor prompts, outputs, and behavior for leakage, abuse, or unexpected actions. Cisco is positioning AI Defense as a control layer that follows AI systems across that full lifecycle.

Cisco also said it is extending Secure Access with AI Access so organizations can discover and control both sanctioned and unsanctioned AI applications, along with shadow AI and shadow agents. That part of the announcement speaks directly to a common enterprise problem. Workers are already experimenting with external AI tools outside approved workflows, and agentic systems raise the stakes because they can act on corporate data and internal systems rather than only answer questions. Discovery and policy enforcement therefore become as important as raw threat detection.

The launch also includes an AI-driven security assistant inside Cisco's security portfolio. The strategic point is that AI is being used on both sides of the problem: enterprises are deploying more agents, and vendors are responding by using AI to help analysts understand, prioritize, and remediate the new classes of risk that follow. Cisco is trying to make security operations faster while also widening the perimeter of what needs to be inspected.

For IT buyers, the announcement is a sign that AI security is moving out of narrow pilot tools and into mainstream network and access architecture. If agentic applications become common in enterprise operations, companies will need policy engines that can identify unofficial AI use, map how models touch data, and restrict risky actions before they spread. Cisco is betting that AI governance will increasingly be enforced through the same connectivity and access stack that enterprises already rely on for security.