Promptective journal / Topic index

tool security

2 journal entries tagged tool security.

Paper, translucent data, woven tool cable and a metal output capsule cross a smoked-glass partition through one cobalt checkpoint

Guide18 July 20264 min read

Designing the AI security boundary

Controls for prompts, data, outputs and tools.

A practical framework for securing LLM applications at the boundary between trusted instructions, untrusted content, model output, and real-world actions.

Read guide

A graphite agent gripper holds a brass tool above a cobalt approval boundary beside an already permitted low-risk tool

Guide2 February 20264 min read

Securing AI agents with tools

Bound permissions and hold consequential actions for approval.

Learn how to reduce prompt injection, tool abuse, and MCP risks with practical controls for LLM agents and their connected systems.

Read guide

All Articles, Guides and News