AI Security Architecture All-in-One

AI Security Architecture All-in-One is an 824-page enterprise architecture handbook for LLMs, AI agents, RAG, MCP, APIs, identity, data security, cloud, zero trust and secure AI platforms. Its organising idea is the Trust Boundary Method: draw the threat model on the architecture diagram itself, find the component that ignores the diagram, and design the boundaries, gateways and controls that make AI systems defensible.Part I opens with the component that ignores the diagram, a threat model drawn on the architecture diagram and the standards that constrain the design. Part II covers discovery and inventory: the AI asset model as a schema for models, indices, tools and identities; finding AI in the estate through egress, IAM, repositories and shadow SaaS; tiering AI systems by the boundaries they cross; and model providers, embedded AI and the supply chain of weights. Part III addresses prioritisation and programme design, beginning with who owns the gateway, ownership and the obligation register.Later parts set out reference architectures for the AI gateway, retrieval and RAG pipelines, agent and tool-calling platforms, MCP integration, identity and secrets for non-human actors, data classification and leakage controls, cloud and platform hardening, zero trust for AI traffic, logging and evidence, and the assurance and governance that keep the architecture honest over time.Written for enterprise and security architects, platform engineers, cloud and identity specialists, CISOs, risk and assurance professionals and technology leaders building or approving AI systems. Part of the Understand, Architect, Attack, Defend, Govern series of enterprise AI security references.Key topics include: AI security architecture; trust boundaries; AI gateway; LLM and agent platforms; RAG and retrieval security; MCP and tool calling; API security; non-human identity; data security and leakage; cloud AI security; zero trust; AI asset inventory; supply chain of models.