QuestionQ59

VerifiedSecuring and governing agentic workflows

You are designing a secure, multiagent framework on Gemini Enterprise Agent Platform. An administrative assistant agent needs access to internal corporate wikis and external communication APIs to synthesize daily updates and draft emails.

You need to establish a security architecture integrated with Agent Gateway to protect against indirect prompt injections and data exfiltration. You want the framework to scan user input prompts and retrieved message content in real time, while enforcing an explicit authorization boundary for critical, state-modifying operations.

What should you do?

Explanation

Model Armor templates integrated with Agent Gateway can screen user prompts and supported tool requests and responses for prompt injection and sensitive data leakage. A human-in-the-loop checkpoint can require explicit approval before a critical state-modifying API operation proceeds. Screening depends on the payload types supported by the gateway integration. (docs.cloud.google.com)

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!