Cisco Debuts New Security Innovations to Protect the Global Agentic AI Workforce

At the RSA Conference 2026, Cisco unveiled a major expansion of its security portfolio specifically designed to govern the emerging ecosystem of autonomous AI agents. These software entities, which go beyond answering questions to performing complex tasks, currently face a significant trust gap in the enterprise. Cisco aims to close this gap by establishing end-to-end security that allows organizations to scale their AI initiatives with confidence.

A central component of the announcement is the extension of Zero Trust Access to the agentic workforce. Through new Duo IAM capabilities, companies can now register AI agents and map them to human owners for full accountability. This ensures that every automated action is traceable and governed by strict, time-bound permissions that prevent unauthorized lateral movement within the network.

Cisco also introduced AI Defense: Explorer Edition, a self-serve tool that allows developers to stress-test their models before they are deployed. This toolkit facilitates dynamic red teaming to uncover vulnerabilities like prompt injection or sensitive data extraction. By providing these resources, Cisco helps technical teams embed security guardrails directly into the agent development lifecycle.

To foster collaboration, Cisco is releasing DefenseClaw, an open-source framework that automates security inventory and scanning for AI skills. This framework is designed to integrate with NVIDIA OpenShell, providing a secure sandbox environment that eliminates manual security steps. Such partnerships are vital as the industry seeks standardized ways to verify the integrity of AI-driven tools.

On the operations side, Splunk is being upgraded with a suite of specialized AI agents to help security teams work at machine speed. These agents handle tasks ranging from malware analysis to incident triage, reducing the burden on human analysts overwhelmed by alert fatigue. These tools effectively use AI to fight the very threats that malicious autonomous agents might create.

Recent data from Cisco shows that while 85 percent of enterprises are experimenting with AI agents, only 5 percent have reached production status. Security remains the primary hurdle for wider adoption, as legacy tools often struggle to monitor non-human identities. Cisco's holistic approach addresses these concerns by treating agents as a new class of digital coworkers.