Learn / Level 7

AI agent security

Agents read, decide and act. Each of those steps is an attack surface, and each lesson maps to a real incident and a Beam control.

1. 01How coding agents workTools, hooks, MCP and the loop between them.8 min
2. 02Prompt injectionInstructions hidden inside data.8 min
3. 03Config poisoningWhen a project file configures the attacker's tool.7 min
4. 04MCP server risksTools you connect are code you trust.8 min
5. 05Plugin and skill trustPinning, hashes and what a bypass costs.7 min
6. 06Credential exfiltration by agentsHow an agent leaks the keys it can read.8 min
7. 07Destructive actions and scopeDeleting the wrong thing, correctly.7 min
8. 08Sandboxing agentsGiving an agent a smaller world.7 min
9. 09Monitoring and judging actionsWhy monitors miss things and what to do about it.8 min
10. 10Policies, audit and RBACMaking agent use governable in a company.8 min