AI and Agentic Security

Interpret the purpose of variables, functions, conditionals, and loops in a short Python script.

Use basic command-line commands to navigate files and run a provided program.

Explain how JSON structures data exchanged between agent components.

Trace a simple HTTP request and response between an agent and a tool.

Store a demonstration secret in an environment variable rather than source code.

Use basic Git commands to inspect and restore versions of demonstration files.

Run a provided agent demonstration in an isolated local environment.

Read application logs to reconstruct the steps taken by a demonstration agent.

Apply confidentiality, integrity, and availability to an agentic AI scenario.

Identify assets, threat actors, entry points, and adverse impacts in an agent workflow.

Mark trust boundaries on an agent data-flow diagram.

Distinguish a vulnerability, threat, exploit, control, and residual risk.

Write a concise threat scenario using actor, action, target, and impact.

Prioritize agentic risks using likelihood and impact.

Map preventive, detective, and corrective controls to a threat scenario.

Explain why probabilistic behavior changes traditional security assumptions.

Distinguish security risks originating from models, applications, tools, data, users, and third parties.

Map common LLM and agentic threats to a recognized security taxonomy.

Explain the confused-deputy problem in an agent that acts for a user.

Identify excessive agency in a proposed agent workflow.

Analyze how multiple individually safe components can create an unsafe agent workflow.

Assign security responsibilities across model providers, application builders, deployers, and users.

Distinguish direct prompt injection from indirect prompt injection.

Explain how untrusted content can compete with trusted instructions.

Demonstrate a benign prompt-injection attack in an isolated example.

Identify prompt-injection paths through web pages, documents, messages, and tool output.

Explain why prompt wording alone cannot reliably prevent prompt injection.

Select layered controls for a prompt-injection scenario.

Assess the residual risk after prompt-injection controls are applied.

Apply least privilege to the tools available to an agent.

Design an allowlist for an agent tool's permitted actions and parameters.

Choose actions that require deterministic checks or human approval.

Explain the difference between user identity, agent identity, and delegated authority.

Identify authorization failures in a multi-user agent scenario.

Select sandboxing and isolation controls for code-executing agents.

Identify security risks in agent-to-tool protocols such as the Model Context Protocol.

Analyze how an attacker could chain tools to exceed an agent's intended authority.

Trace how source content moves through a retrieval-augmented generation pipeline.

Identify poisoning and exfiltration opportunities in a retrieval pipeline.

Select access controls for documents, embeddings, and vector stores.

Explain how persistent memory can create cross-session security risk.

Design validation rules for information written to agent memory.

Assess data minimization needs for prompts, logs, memory, and tool output.

Identify supply-chain risks in models, plugins, packages, datasets, and hosted services.

Use provenance information to judge whether an agent dependency is trustworthy.

Construct a defense-in-depth architecture for a representative enterprise agent.

Separate trusted control logic from probabilistic model output in an architecture diagram.

Place input validation, output validation, and policy enforcement at appropriate boundaries.

Select fail-safe behavior for unavailable or uncertain security controls.

Write security requirements for one agent capability.

Add agentic security activities to each phase of a development lifecycle.

Define deployment gates based on security evidence and residual risk.

Review an agent architecture for missing controls and unsafe assumptions.

Derive a security test case from an agent threat scenario.

Distinguish functional evaluation, safety evaluation, penetration testing, and red teaming.

Create benign adversarial inputs for prompt, retrieval, memory, and tool-use tests.

Define pass criteria for an agent security test.

Execute a structured security test against an isolated demonstration agent.

Interpret attack-success rate, false-positive rate, and task-utility measurements.

Document a security finding with evidence, impact, and remediation.

Apply authorization, scope, and data-handling rules to an AI red-team exercise.

Select logs needed to investigate prompts, retrieval, tool calls, approvals, and policy decisions.

Define indicators of suspicious agent behavior.

Differentiate an agent attack from an ordinary model or application failure.

Triage a suspected agentic security incident using impact and urgency.

Choose containment actions for compromised tools, credentials, memory, or data sources.

Draft an incident playbook for one agentic attack scenario.

Conduct a tabletop walkthrough of an agentic security incident.

Translate incident lessons into control and training improvements.

Map an agentic AI use case to the Govern, Map, Measure, and Manage functions of the NIST AI Risk Management Framework.

Classify an agent use case by autonomy, data sensitivity, external exposure, and impact.

Define accountable roles for agent ownership, security, privacy, compliance, and operations.

Create a concise risk register entry for an agentic AI risk.

Evaluate a third-party agent or model using a structured security questionnaire.

Select evidence needed to assure leaders that an agent control is operating.

Define acceptable-use boundaries for employees using agentic AI.

Explain how legal, privacy, safety, and security governance interact in an agent deployment.

Segment organizational audiences by the agentic security decisions they make.

Write measurable learning objectives for an agentic security training session.

Convert a technical agent threat into a role-relevant workplace scenario.

Design a short demonstration that safely illustrates one agentic attack and its controls.

Create a facilitator guide for an agentic security tabletop exercise.

Build a role-based learning pathway for builders, testers, leaders, and general users.

Define adoption and behavior metrics for an agentic security enablement program.

Present a capstone enablement plan linking threats, controls, governance, and incident readiness.