Скачать или смотреть Stop Your AI Agents from Going Rogue: Enterprise Security for OpenClaw Guide

2026 is the year of the AI agent, and OpenClaw is leading the charge. But as enterprises adopt agentic workflows, security is the #1 hurdle. In this video, John Cuneo (Global AI Solutions Engineer at Cisco) breaks down how to move beyond basic application security to harden the system and the model layers.

We dive deep into a local deployment using vLLM and Qwen-3 to demonstrate real-time monitoring and enforcement against modern AI threat vectors.

What You’ll Learn:
System-Level Security: Using eBPF and Tetragon for kernel-level visibility into file operations and network calls.

Skill Scanning: How to use the Cisco AI Defense Skill Scanner to vet third-party skills from Claw Hub for malicious code and prompt injection.

Model Guardrails: Implementing Cisco AI Defense to block harmful prompts, jailbreaks, and PII leaks in real-time.

Infrastructure Hardening: Why sandboxing isn't enough and how to implement a network "kill switch" for rogue agents.

Resources & Links:
My site: https://cuneocode.com
Tetragon (Isovalent): https://tetragon.io
Cisco AI Defense Skill Scanner: https://github.com/cisco-ai-defense/s...
OpenClaw Project: https://openclaw.ai
AI Infrastructure Planner: https://ai-infrastructure-planner.com

Video Chapters:
0:00 - Why Agent Security Matters in 2026
1:15 - Meet "Lydia": The Local OpenClaw Demo
2:30 - System vs. Model Threat Taxonomy
3:45 - eBPF & Tetragon: Kernel-Level Defense
5:20 - Live Demo: Monitoring File Deletion & Network Lockout
7:10 - Skill Scanner: Securing the AI Supply Chain
9:15 - Blocking Harmful Prompts with Cisco AI Defense
10:45 - Scaling Security for the Enterprise

Key Tech Stack:
Agent Framework: OpenClaw
Inference: vLLM on NVIDIA DGX Spark
Observability & System Security: Tetragon (eBPF)
Model Security: Cisco AI Defense

#AIAgents #OpenClaw #CyberSecurity #eBPF #Cisco #AIInfrastructure #LLMSecurity