[Lab] Building with Google Antigravity → https://goo.gle/4kXTWpN
[Lab] Securing a Multi-Agent System → https://goo.gle/4aDl2zb
[Lab] Build a Secure Agent with Model Armor and Identity → https://goo.gle/4cffP1z
Build a production agent using vibe coding with Antigravity. Learn how to implement critical security checkpoints using Model Armor callbacks. Working in small, test-driven phases. Aron shares how to intercept a user’s prompt before it hits the model to block injection attacks, and how to scrub the model’s response before it ever reaches the user to prevent data leakage.
Chapters:
0:00 – Security checkpoints in production agents
0:38 – The Plan: Integrating Model Armor with ADK callbacks
0:51 – Step 1: Input guardrails and prompt injection defense
0:58- Red Phase: Simulating malicious user input
1:15 – Green Phase: Implementing before_model_callback
1:42 – Step 2: Output validation and data leakage prevention
1:55 – Red Phase: Mocking sensitive model responses
2:12 – Green Phase: Sanitizing output with after_model_callback
2:26 – Refactoring and consolidating security policies
More resources:
Immersive Stream for XR documentation → https://goo.gle/4aDWshE
🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech
#GoogleCloud
Speaker: Aron Eidelman
Products Mentioned: Google Cloud Security, Gemini CLI, Antigravity
