As AI agents move from passive chatbots to autonomous actors that can modify files, run database commands, and change firewall rules, enterprises face a critical security gap: 85% have agent pilots underway but only 5% have moved them to production. Cisco is addressing this trust deficit across three fronts: protecting agents from manipulation (prompt injection, adversarial attacks) via tools like AI Defense Explorer Edition, Agent Runtime SDK, and the open-source DefenseClaw framework; protecting systems from agents through Zero Trust identity controls in Duo IAM with time-bound, task-specific permissions; and enabling machine-speed threat detection and response via new Splunk capabilities including Exposure Analytics, Detection Studio, and Federated Search. The core argument is that robust security is not a brake on AI adoption but the enabler that lets organizations move from pilots to production.
Table of contents
From Answering to ActingThe Gap Between Pilots and ProductionProtecting agents from the world means ensuring agents can’t be manipulated by bad actors.Protecting the world from agents is an identity and access problem.Finally, we have to detect and respond to security threats and incidents at machine speed .Security is the AcceleratorSort: