Series

Designing Secure AI Agents

A practical series on building AI agents that are secure by design, covering trust boundaries, prompt injection defenses, scoped tool access, approval gates, and audit architecture. Each post stands alone; together they form a complete playbook.

5 posts. Read in order or jump to any part.

AI Agent Architecture: The Trust Boundary Model

A practical architecture model for secure AI agents: separate trusted instructions, untrusted data, tool access, external actions, memory, identity, approval gates, and audit logs.

AICybersecuritySeries

May 23, 2026Read

AI Agent Memory vs State: What Should Be Remembered, Stored, or Recomputed?

AI agent memory vs state explained for enterprise teams: what to remember, store, recompute, audit, and secure before production.

AICybersecuritySeries

May 30, 2026Read

Tool Output Is Not Instruction: A Core Rule for Secure AI Agents

Tool output is data, not instruction. How secure AI agents read untrusted emails, webpages, and API responses without losing control.

AICybersecuritySeries

June 13, 2026Read

Secure Architecture for AI Agents That Read Email, Documents, and Webpages

Design secure AI agents that read email, documents, and webpages using content boundaries, least-privilege connectors, RAG controls, action gates, and audit logs.

AICybersecuritySeries

June 20, 2026Read

AI Agent Prompt Injection Risk Scorecard: How to Assess Agents Before Production

Score AI agent prompt injection risk: source exposure, data sensitivity, tool access, approval gates, and runtime controls. A 12-dimension enterprise assessment.

AICybersecuritySeries

July 19, 2026Read