March 23, 2026
Cisco Security Updates Address ‘Top Barrier’ to AI Agent Adoption
Security concerns remain one of the most significant obstacles standing between AI agents and widespread enterprise adoption. Cisco has long championed a future where AI operates as a digital co-worker, but for that vision to take hold, organisations must first be able to trust the technology. At RSA Conference 2026, the company unveiled a suite of security innovations designed to “remove a top barrier to agent adoption”. The company attributes each update to one of three ambitions: protect the world from agents, protect agents from the world, and identifying threats and acting on them at the speed AI itself operates.
The Adoption Gap
A Cisco survey of major enterprise customers found that 85% were experimenting with AI agents, yet only 5% had taken agents through to production. The gap between interest and deployment reflects a broader anxiety that deploying autonomous software at scale introduces risks that traditional security tooling was never built to handle. We already saw in the first half of 2025, more than 8,000 global data breaches, exposing an estimated 345 million records.
Jeetu Patel, President and Chief Product Officer at Cisco, believes security can provide the answer: “AI agents aren’t just making existing work faster; they’re a new workforce of co-workers that dramatically expand what organisations can accomplish. The only limit is imagination, and security teams are the key to unlocking this opportunity by making the agentic workforce safe enough to trust.”
‘Protecting the World from Agents’
Cisco’s first pillar addresses the risk that agents act beyond their intended scope. The company is extending Zero Trust Access to AI agents, with new capabilities in Duo allowing organisations to register agents, tie each one to a named human owner, and assign fine-grained, time-bound permissions. All agent traffic is routed through a model context protocol (MCP) gateway to eliminate blind spots. The approach is informed by Cisco’s 2025 Talos Year in Review, which found that attackers have increasingly concentrated on the infrastructure responsible for user authentication, access control, and trust between systems. With internal AI behaviour emerging as a security concern, the identity-binding approach aims to establish clear accountability before agents are set to work.
‘Protecting Agents from the World’
The second pillar focuses on agents’ vulnerability to manipulation. Concerns around data exposure and privacy have shadowed AI adoption, and Cisco is seeking to help tackle these at the build stage with AI Defense: Explorer Edition. The self-service tool that lets developers test models and applications before deployment to look for vulnerabilities to prompt injection, jailbreaks, and other adversarial inputs.
The Agent Runtime SDK embeds policy enforcement directly into agent workflows at build time, compatible with major platforms such as AWS Bedrock AgentCore, Google Vertex Agent Builder, and Azure AI Foundry. Cisco is also publishing the LLM Security Leaderboard, a public resource for evaluating how models perform against malicious prompts and manipulation attempts. The open-source DefenseClaw framework automates security scanning and asset inventory across agent deployments, with planned integration into NVIDIA’s OpenShell for sandbox-level enforcement.
Rapid Detection and Response
Splunk, part of Cisco’s security portfolio, is adding a series of AI-powered capabilities aimed at shifting the security operations centre (SOC) from reactive to proactive. Exposure analytics provide real-time asset and user visibility and Detection Studio streamlines the full detection engineering lifecycle. Meanwhile, a suite of specialist AI agents, including Triage, Guided Response, and Malware Threat Reversing agents, automate key SOC workflows. The Talos Year in Review found that vulnerabilities are now being exploited at near-instant speed, likely accelerated by AI-generated exploit kits, making the case that defenders need to operate at the same pace.
A Foundation for Trust
These updates represent a substantial attempt to address the security questions that have kept many organisations from bridging the readiness gap between AI experimentation and full deployment. Whether they are enough to shift the 95% of enterprises yet to move agents into production remains to be seen, but Cisco is right to be focussing on agentic security as it is undoubtedly a central area of concern to these businesses.
