Learn how to securely build AI agents that interact with user-specific data in a database. This video outlines a high-level, layered approach to securing agents, starting with Identity and Access Management (IAM) for controlling agent access. We then explore the critical intermediary pattern for isolating agents from managing credentials and direct data access, using secure Tools. Finally, discover how Model Armor and Sensitive Data Protection (SDP) add a final layer of defense by inspecting and redacting sensitive information in conversational data.
Chapters:
0:00 - Introduction: Secure Agents and User Data
0:19 - Controlling Agent Management and Interaction with IAM
1:13 - Isolating Agent Credentials: The Intermediary Pattern
1:35 - How Agent Tools Access User-Specific Data Securely
2:54 - Handling Third-Party API Keys without Agent Exposure
3:11 - Protecting Conversational Data: Model Armor and SDP
3:27 - The Three Layers of Agent Security
Resources:
Previous videos
- Secure AI: De-identifying data with SDP → http://goo.gle/4htbmcf
- Secure ADK agents with Secret Manager → http://goo.gle/4oEnFF3
- Agent Sessions and Tool Authentication → http://goo.gle/4oErjid
Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech
#GoogleCloud #IAM #SDP #AIAgent
Speakers: Aron Eidelman
Products Mentioned: Google Cloud Security