How to secure your AI Agents: A Technical Deep-dive

Опубликовано 3 декабря 2025, 20:00

AI agents introduce unique security challenges like prompt injection, data leakage, and excessive agency. This workshop breaks down the OWASP LLM Top 10 and demonstrates how to implement robust security controls using Model Armor and the Agent Development Kit (ADK). Learn how to filter malicious inputs, redact sensitive data from outputs, and manage authentication patterns to build secure and reliable agentic workflows.

Resources:
Explore the Agent Development Kit → google.github.io/adk-docs
Subscribe to Google for Developers → goo.gle/developers

Chapters:
0:00 - Introduction to agent security
1:32 - OWASP LLM top 10 vulnerabilities
2:56 - Sensitive information disclosure
4:10 - Excessive agency risks
5:11 - Input filtering with Model Armor
6:26 - Demo: blocking prompt injection
10:18 - Output filtering and PII redaction
11:09 - Demo: redacting sensitive data
12:43 - Authentication and authorization patterns
13:44 - Agent-to-tool authentication flow
16:13 - Managing API keys with Secret Manager
18:05 - Security for A2A and MCP protocols
19:22 - Supply chain security measures
22:25 -Protecting agent access and integrity

Subscribe to Google for Developers → goo.gle/developers

Speaker: Sita Lakshmi
Products Mentioned: Agent Development Kit (ADK)

#AIAgents #ADK #AgenticSecurity