Agentic AI Security Lab with Apoorva Sharma

AppyGuru’s Agentic AI Security Lab is an original, defensive course pack for security, platform, AI product, GRC, and incident-response teams that need to rehearse the unusual failure modes of tool-using agents. It focuses on safe, authorized learning: understand what an agent can read and change, model trust boundaries, practice controlled scenarios in staging, test detection and containment, and leave with owners and retest dates. It is not a pentest license, exploit collection, or guarantee of security.

The Lab Guidebook explains why autonomy expands blast radius: indirect instructions can enter through tickets or web content, a confused deputy can act through an over-scoped identity, memory can preserve a false instruction, and a planner can loop into cost or operational harm. It provides a mandatory safety brief, learning objectives, facilitator flow, boundary sketch, agent-specific abuse patterns, scenario stations, runbook drill, success criteria, knowledge checks, detailed prompts, evidence habits, scoring guidance, authorization gates, and after-action structure.

File contents: Lab Guidebook; Threat-Modeling Worksheet; Red-Team Scenario Pack with facilitator-ready controlled tabletop scenarios; Word Incident Runbooks; PowerPoint tabletop deck; Excel tracker with scenario scores, detection log, control backlog, lessons, threat register, test cases, authorization log, evidence, status, and retests.

Recommended use: obtain written authorization; use synthetic data and staging; identify kill switch and credential revocation; baseline normal behavior; run one controlled scenario at a time; score detection and response; preserve the timeline; file three control improvements; and retest.

Educational scope only: AppyGuru materials do not authorize access to systems you do not own or lack permission to test, and do not replace professional security review, counsel, or an organization’s response plan.