Learn
AI agent security and governance, explained plainly.
AI agent security is the set of controls that decide what an AI agent may do, ask a person before risky actions, and prove afterwards what happened. These guides explain each control, the standards behind it, and how to put it in place.
Updated 31 October 2026.
Concepts
What is an MCP gateway?
What sits between agents and MCP servers, and what it should enforce.
What are guardian agents?
Gartner's term for the layer that oversees other agents, and what it needs to do.
AI agent governance
Identity, least privilege, runtime enforcement, oversight and evidence, as a checklist.
Human-in-the-loop approval
When to hold an agent action for a person, and how four-eyes approval works.
The AI agent kill switch
Stop the process, revoke the credentials or block at the gateway: which one works.
Sovereign and air-gapped agents
Running and governing agents in your own jurisdiction, on your own keys.
Agent identity and least privilege
Short-lived, scoped identities that say which agent acts for which person.
Problems and how to solve them
Stop an AI agent deleting production
What went wrong in real incidents, and the layered fix.
Audit what an AI agent did
What to record, how hash chaining works, and how a third party verifies it.
Prompt injection in tool calls
Why detection alone fails, and why you constrain actions instead.
Is MCP safe?
The real MCP security risks, what the spec requires, and the controls for each.
Stop agents leaking data
Egress control, DLP and taint: one way out, inspected.
Standards and regulation
OWASP Top 10 for Agentic Applications
ASI01 to ASI10, each mapped to a control.
The EU AI Act and AI agents
Record-keeping, human oversight and the current timeline.
ISO/IEC 42001 and NIST AI RMF for agents
Turning management-system controls into runtime evidence.
Try the controls yourself.
Run MAQPNA on a laptop and watch a policy allow, hold and deny real tool calls.
curl -fsSL https://maqpna.com/install.sh | shirm https://maqpna.com/install.ps1 | iex