Numbered references support the corresponding discussions. Architecture diagrams and fictional scenarios are original teaching constructions. They illustrate design choices and are not official NIST, OWASP or MCP diagrams.
[1] NIST AI RMF 1.0. Artificial Intelligence Risk Management Framework, NIST AI 100-1, January 2023. Basis for the Govern, Map, Measure and Manage discussion. The framework addresses trustworthiness across the lifecycle. NIST states that AI RMF 1.0 is under revision as of this edition. Check the current framework when applying it.
[2] NIST Generative AI Profile. NIST AI 600-1, 26 July 2024. Companion to AI RMF for generative AI, including risks beyond malicious attacks. Used to check privacy, information integrity, human oversight and lifecycle coverage.
[3] OWASP GenAI LLM Top 10 2026. Published 3 August 2026. Used for the application-security review. Risk identifiers differ between editions, so do not mix them.
[4] OWASP Top 10 for Agentic Applications 2026. Released 9 December 2025. “2026” is the edition name, not the publication year. Used for agent authority, memory, tool misuse and failure scenarios. The separate Agent Control Standard, published 1 September 2026, is a newly released implementation resource that this guide’s architecture does not require.
[5–7] Model Context Protocol. Specification and architecture for version 2026-07-28, plus its authorization requirements. Published 28 July 2026 as a final release with a stateless protocol core. Used to verify host/client/server roles, primitives, the stateless core and token handling. Use version-matched SDK documentation when implementing.
This guide does not claim compliance, certification or exhaustive threat coverage. Apply the full sources to the actual data, users, jurisdiction and deployment. Recheck specifications and product controls before implementation.