AI Guardrails Are Operating Controls, Not Prompt Instructions
A practical baseline for governing AI authority, data, tools, consequential actions, evidence, and lifecycle risk without mistaking a system prompt for a security boundary.
ai-security
Latest
Inspect Both Sides of the LLM Conversation
A practical policy and configuration guide for detecting prompt injection and jailbreak attempts before inference, then blocking unsafe content and sensitive-data disclosure before model output reaches a user, tool, or application.
Do AI Firewall Sidecars Make Sense with Vendor-Hosted LLMs?
A reader of the Cybersecurity Architect's Handbook Second Edition recently asked whether the Kubernetes sidecar pattern discussed in Chapter 13 applies only when an organization hosts its own large language model. The short answer is no.
Vaultwarden and HashiCorp Vault: Different Jobs in the Same Secrets Architecture
Yesterday I wrote about moving autonomous-agent secrets out of a flat .env file. Why I kept Vaultwarden for people and recovery, added HashiCorp Vault for workloads, and treated identity, audit, PKI, and tested recovery as part of the deployment rather than follow-up work.
Moving Autonomous Agent Secrets Out of .env
For a long time, my autonomous agent found credentials the same way many applications do. Why I replaced a flat environment file with scoped Vaultwarden access, short-lived agent sessions, and a verified audit trail that now reaches Graylog and Wazuh.
The Moment AI Stops Being a Tool: Why Autonomy Is the Risk Multiplier
A tool's failure is advisory because a human still reviews the output before anything happens. An agent's failure is operational because the bad output is the action...