Research snapshot · 2 October 2026

AI Agent Security

From influence to control.

Thirteen lessons, in order. Each one defines the words, explains the mechanism, then asks you to make one decision. A wrong answer shows what still happens.

1. 01The security modelOpen
2. 02The October 2026 landscapeOpen
3. 03The attack atlasOpen
4. 04Attack labOpen
5. 05A defensible architectureOpen
6. 06Identity, authorization, and delegationOpen
7. 07MCP, tools, and skillsOpen
8. 08Data, retrieval, and memoryOpen
9. 09Execution and computer useOpen
10. 10Testing that proves somethingOpen
11. 11Operations, governance, and responseOpen
12. 12Build a defensible briefOpen
13. 13Reading roomOpen

The brief and the drill sit beside the lessons. They are not graded.