In Production: The Defense Stack Around a Served Model

The domain’s production capstone: defense in depth around a model that cannot defend itself — provider filters, app-side controls, and monitoring.

Content last verified 2026-09.

Lessons

1. Defense in Depth
2. The Cloud Safety Surfaces
3. Monitoring and Response

Sources

• OWASP Top 10 for LLM Applications (2025)
• Amazon Bedrock User Guide — Guardrails
• Azure AI Foundry — Content filtering (Microsoft Learn)
• Vertex AI — Configure safety and content filters
• Greshake et al. (2023) — Not what you’ve signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection