Certified Offensive AI Security Professional (C|OASP) — Lesson 3: Data Poisoning, Model Extraction, Evasion, Supply Chain, and Monitoring

Learning Objectives

• Explain the roles, services, data, settings, dependencies, and decision criteria governing data-poisoning risk and model-extraction risk.
• Implement and validate adversarial evasion, AI supply chain, and security monitoring using controlled evidence and clear acceptance criteria.
• Evaluate the condition where analysts reach the wrong conclusion because security monitoring evidence conflicts with model-extraction risk context and the baseline was not preserved and select a proportionate corrective response.

Detailed Topics

• data-poisoning risk applied to the course scenario, expected behavior, dependencies, controls, risks, and evidence.
• model-extraction risk applied to the course scenario, expected behavior, dependencies, controls, risks, and evidence.
• adversarial evasion applied to the course scenario, expected behavior, dependencies, controls, risks, and evidence.
• AI supply chain applied to the course scenario, expected behavior, dependencies, controls, risks, and evidence.
• security monitoring applied to the course scenario, expected behavior, dependencies, controls, risks, and evidence.

Instructor Demonstration

The instructor traces an authorized AI red team evaluating model and application weaknesses safely, preserving evidence, and strengthening defensive controls. The walkthrough connects tools, configurations or analysis logic, system messages, decisions, and evidence captured in the AI red-team charter, attack-scenario evidence, defensive-control assessment, and retest report. It also tests the exception where analysts reach the wrong conclusion because security monitoring evidence conflicts with model-extraction risk context and the baseline was not preserved.

Hands-On Exercise

Complete a controlled, authorized EC-Council exercise covering Data Poisoning, Model Extraction, Evasion, Supply Chain, and Monitoring and preserve the verified result in the AI red-team charter, attack-scenario evidence, defensive-control assessment, and retest report.

1. Confirm scope, authorization, prerequisites, versions, identities, dependencies, and acceptance criteria.
2. Execute the workflow while recording settings, outputs, events, decisions, and system messages.
3. Test normal behavior, one boundary condition, and the stated exception without targeting real third-party systems.
4. Document corrections, evidence, ownership, retest criteria, rollback or recovery, and handover actions.

Scenario Analysis

• Identify which object, service, policy, setting, or sequence most strongly influences data-poisoning risk.
• Compare two defensible responses and explain technical, security, operational, legal, and support trade-offs.
• Determine how the AI red-team charter, attack-scenario evidence, defensive-control assessment, and retest report should be approved, retained, retested, and updated.

Troubleshooting Guidance

Start where expected behavior first differs from the observed result. Check authorization, prerequisites, identity, configuration order, data quality, version compatibility, connectivity, status, resource limits, and evidence separately. Treat “analysts reach the wrong conclusion because security monitoring evidence conflicts with model-extraction risk context and the baseline was not preserved” as a hypothesis until logs, settings, captured outputs, and retest outcomes confirm the cause.

Validation Checklist

• Required components, identities, permissions, settings, safety controls, and dependencies are present.
• The AI red-team charter, attack-scenario evidence, defensive-control assessment, and retest report connects each conclusion to objective, reviewable evidence.
• Normal processing, a boundary condition, dependency behavior, and the stated exception were tested.
• Corrections, ownership, retest criteria, rollback or recovery, documentation, and operational handover are complete.

Lesson Outcome

Learners can apply Data Poisoning, Model Extraction, Evasion, Supply Chain, and Monitoring, explain the resulting security, governance, business, or operational behavior, and defend the outcome through the AI red-team charter, attack-scenario evidence, defensive-control assessment, and retest report.