QuestionQ58

VerifiedSecuring and governing agentic workflows

You are designing an enterprise multiagent procurement system in Google Cloud. An orchestrator agent processes vendor purchase orders, retrieves contract terms, and invokes downstream tools. Your organization’s security policies require you to stop adversarial prompt injection in vendor documents, and high-risk financial actions must receive explicit compliance officer authorization before execution.

How should you meet these requirements?

Explanation

Model Armor security templates on Agent Gateway can inspect supported agent and tool traffic for prompt injection, including malicious instructions in retrieved content. Human-in-the-loop approval controls can pause high-risk financial tool calls until a compliance officer authorizes them, before execution. (docs.cloud.google.com)

Gateway integration does not screen document or file uploads such as PDFs directly; those require direct Model Armor document screening. (docs.cloud.google.com)

Learn more

Community Discussion

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