Cisco Debuts DefenseClaw: First Open-Source Security Framework for AI Agents

Cisco launched DefenseClaw, an open-source framework scanning AI agents for vulnerabilities and regulating their resource interactions, integrating with Nvidia OpenShell.

Cisco Debuts DefenseClaw for AI Agent Security

Cisco launched DefenseClaw on March 23, the first open-source security framework specifically designed for AI agents. Features: vulnerability scanning (prompt injection, privilege escalation, data leakage), behavior monitoring (API/database/filesystem interactions), and Nvidia OpenShell integration for GPU pipeline security checks.

Plugin-based architecture allows customizable security rules. Complements Kubescape 4.0 - Kubescape covers K8s-level agent policies while DefenseClaw covers code and interaction-level scanning.

Cisco's entry signals AI agent security moving from academic discussion to industrial stage. Open-source approach aims to catalyze a developer ecosystem around agent security standards.