Secure distributed multi-agent systems
summary
A demo web app deploys four Cloud Run agents (orchestrator, researcher, judge, content builder) that create courses and shows how to stop attacks before wasting resources: requests pass through Model Armor and Sensitive Data Protection templates that inspect and optionally redact or block content, flagging prompt injection, malicious URLs, jailbreak attempts and >200 info types (credit cards validated with Luhn, API keys, SSNs). Logs, reversible hashing for troubleshooting, SDK/Agent Platform integrations and a cold‑start end‑to‑end lab prove the defenses work.
AI-written summary · about
chapters (16)
- 0:00Multi-agent security intro
- 1:07What is MAS security
- 3:39Prompt injection risks
- 6:51Why guardrails alone aren't enough
- 10:33Live demo: Model Armor
- 15:16Multi-agent architecture
- 21:22Model Armor code walkthrough
- 27:18Testing Model Armor
- 30:04Configuring templates
- 39:13Sensitive data protection
- 45:16Deployment modes
- 49:42Reinforcement learning attacks
- 53:40Supervision in workflows
- 56:35Agent-to-agent security
- 1:00:51AI agents in IT ops
- 1:05:55Q&A
related talks
11mBuilding Secure Backend Services with AI Agents
15mConquering Agent Chaos
24mAI Agent Swarms Are Amazing
28mAgent Runtime Security: Enforce, Attest, Decide
22mBuilding Secure Services with AI Agents: Patterns From Real-World Engineering
37mAGENTS OF S.E.A.L.E.D: AI Agentic Cybersecurity Framework
27mKeeping Your Agents in Check: Layered Security for Agentic Platforms in Production
42mSecuring AI Agents...