Practical lesson

Exercises AI Security

Practise deliberately with small tasks that produce observable evidence of improvement.

The idea in one minute

AI security applies security engineering and threat reasoning to systems whose behavior depends partly on machine-learning or generative models. It covers conventional application and cloud security plus AI-specific attack surfaces such as direct and indirect prompt injection, malicious retrieved content, model or data poisoning, sensitive-information disclosure, insecure output handling, vector and embedding weaknesses, system-prompt leakage, excessive agency, model theft, and attacks on agent goals, memory, tools, identities, or inter-agent communication. Strong practitioners avoid treating the model as a trusted security boundary. They design least privilege around tools and data, validate actions outside the model, isolate untrusted content, protect credentials, monitor behavior, test adversarial cases, and build containment and recovery for failures that cannot be completely prevented.

This capability connects directly with AI Governance, AI Risk Management, Cybersecurity Awareness. Open those concepts when the lesson depends on them rather than treating AI Security as an isolated ability.

Beginner exercises

  1. 1.Draw the trust boundaries around one AI application
  2. 2.Identify which inputs are untrusted and which actions could cause harm
  3. 3.Learn why a system prompt is not an authorization control
  4. 4.Map one AI application against relevant OWASP GenAI risks

Applied exercises

  1. 1.Threat-model direct and indirect prompt injection in a RAG workflow
  2. 2.Move authorization checks outside the model and test bypass attempts
  3. 3.Reduce an agent's tool and credential permissions to least privilege
  4. 4.Create adversarial test cases for data leakage and unsafe actions

Measure your progress

  1. 1.Track exploitable attack paths, percentage of high-impact actions protected by deterministic authorization, least-privilege coverage, secret exposure findings, adversarial test pass rates, detection time, containment time, and recurrence of known weaknesses. Mature security is demonstrated by reduced blast radius and repeatable evidence, not claims that prompt injection has been eliminated.

Build the surrounding skill cluster

AI GovernanceThe ability to establish decision rights, policies, evidence, oversight, and accountability so AI is selected, built, used, monitored, and retired in ways consistent with organizational goals and acceptable risk.AI Risk ManagementThe ability to identify, analyze, prioritize, treat, monitor, and communicate risks created or amplified by AI across its technical, human, business, and societal context.Cybersecurity AwarenessUnderstanding cybersecurity threats and best practices for digital safety.Retrieval-Augmented Generation (RAG)The ability to design and evaluate AI systems that retrieve relevant external knowledge at query time and supply it as grounding context for generated answers or actions.AI Evaluation & BenchmarkingThe ability to design repeatable tests that measure whether an AI system is accurate, useful, safe, reliable, efficient, and fit for a specific real-world purpose.

Keep building this skill

Return to the complete guide for career context, evidence, related skills, practice and progression.

Open the complete AI Security guide →