You finished AG03. Here's what you now understand:
✓What Identity and Privilege Abuse is — and why the attribution gap makes traditional least-privilege enforcement impossible in agentic systems
✓Five vulnerability types: Un-scoped Privilege Inheritance, Memory-Based Retention, Cross-Agent Trust, TOCTOU Authorization Drift, and Synthetic Identity Injection
✓Seven real attack scenarios — from delegated finance agent abuse to forged agent personas in internal registries
✓Why standard IAM systems break when applied to agents: blurred attribution, implicit credential sharing, persistent state, and recursive delegation
✓Nine mitigations — scoped tokens, memory isolation, per-action authorization, human-in-the-loop, intent binding, managed identities, and detection
✓Which mitigations stop which attacks — and why MIT 01 + MIT 07 provide the widest coverage per the matrix