Cisco on the 23rd (local time) unveiled security innovations aimed at the agentic AI ecosystem at RSA Conference 2026, the world's largest cybersecurity exhibition, held in San Francisco.
The key is to grant AI agents trust-based identity and apply zero-trust access control to embed security at every stage before and after deployment. Cisco plans to help corporations move AI agents, which are stuck at the pilot level, into actual operations.
According to Cisco, 85% of major enterprise customers are testing AI agent pilots, but the actual commercialization rate is only 5%. Cisco pointed to security concerns as the cause of this gap and said it combines Duo IAM and Secure Access to provide agent identity management, tool visibility, and a short-term, granular authorization model. It also supports multi-turn red team testing, validation against prompt injection and jailbreaks, and security reporting through the AI Defense Explorer Edition.
On the security operations side, Cisco presented a blueprint for an agentic SOC composed of exposure analytics, Detection Studio, federated search, and various specialized AI agents, with Splunk at the forefront. Cisco said it will automate security procedures in the development phase through integration with the open-source framework Defenseclaw and Nvidia OpenShell, and implement detection and response at machine speed in the operations phase.
Jitu Patel, Cisco president and chief product officer, said, "Security teams that can make the agentic workforce safe enough to trust are the key to turning this opportunity into reality."