Papers · v1.0 · 28 pages

AI Gateway Security Patterns

Prompt injection controls, tool governance, and audit trails for multi-model enterprise gateways.

Architecture diagram context
Reference deployment topology — gateway, inference, and data plane.

Threat model

  • Prompt injection and jailbreaks at ingress
  • Tool abuse via MCP and function calling
  • Cost attacks via unbounded agent loops

Illustration

Place policy enforcement before the model — not after the response.

Knowledge Graph

Related

AI is easy. Running it in production is hard.