Skip to content

AI Security reference architecture

AI security architecture

A security layer between applications and models covering prompts, data, agents, MCP and policy.

AI security layerinspection point

Design principles

  • Treat model input and output as untrusted
  • Enforce policy outside the model
  • Least privilege for every agent and tool
  • Log every AI interaction

Components

ComponentRole
AI gatewayRouting, keys, quotas, logging
Prompt securityInjection and jailbreak detection
Data protectionClassification and redaction
Policy engineCentral allow and deny decisions

Adapt this to your environment

We'll help you apply this pattern to your constraints and existing estate.