Guides to governing AI agents
Short, practical guides for people putting AI agents near real tools: what to control, how to keep a human in the loop, and how to prove afterwards what happened.
Human in the loop for AI agents: approve risky actions before they run
How to put a human approval step between an AI agent and the actions that matter: what to hold, what to let run, and how to keep approvals from becoming noise.
AI agent governance: a practical checklist
A working checklist for governing AI agents: identity, least privilege, approvals, limits, audit and an off switch. Written for teams putting LLM agents near real tools.
What is an MCP gateway, and when do you need one?
An MCP gateway sits between AI agents and MCP servers. What it does, how it differs from connecting agents directly, and what to look for in one.
MCP security: tool poisoning, prompt injection and how to defend an agent
How MCP servers can attack the agents that use them: poisoned tool descriptions, changed tools, and injected instructions. Practical defences that do not rely on the model.
An audit trail for AI agents: what to record and how to prove it
What an audit log for AI agents should contain, why ordinary application logs are not enough, and how signed receipts make the record verifiable.
Guardrails for LLM agents: levels, limits and an off switch
Guardrails that work on LLM agents because they sit outside the model: permission levels, rate and daily limits, automatic pauses and a kill switch.
Keeping autonomous AI agents under control
Autonomous agents act without waiting for you. How to give them room to work while keeping the decisions that matter: scope, approvals, notification and review.
Enterprise AI agents: approvals, roles and the two-person rule
How teams run AI agents safely: separate owners from approvers, require two people for irreversible actions, and keep evidence for audit and compliance reviews.
Add an approval step to Claude, ChatGPT and Cursor
How to make Claude, ChatGPT, Cursor and other MCP-capable agents ask before they send, share or delete, by connecting them through one address.
New to the vocabulary? Start with the glossary.