Skip to main content

Series

AI under lock - practical AI security

AI now has access to our terminals, files, repositories and databases, and most of us run it in auto mode without reviewing every action. This series shows how to secure that work with mechanisms that live outside the model, instead of hoping the model follows instructions. We start with everyday work in Claude Code: preventing leaks of tokens, passwords and personal data, and cutting off access to specific folders (like iCloud Drive) and applications (like Docker) using permissions, hooks and sandboxing. Then we move on to agents you build yourself: input and output guardrails, separating data from instructions in prompts, database access only through tools whose connection lives in code, and automated prompt testing. Every article comes with concrete configuration and code you can copy into your own project.

Posts in this series