ISO/IEC 27090:2026 Cybersecurity — Artificial Intelligence

Prepare with confidence for your ISO/IEC 27090:2026 (Cybersecurity — Artificial Intelligence) certification and technical assessments using this comprehensive practice test course. Designed to simulate the style, structure, and difficulty of professional AI cybersecurity certification exams, this course includes carefully crafted multiple-choice questions (QCM) with detailed explanations to help you strengthen your knowledge and identify areas for improvement.

Whether you are preparing for an AI Security certification, implementing AI-specific security controls alongside an Information Security Management System (ISO/IEC 27001) or AI Management System (ISO/IEC 42001), or looking to validate your understanding of ISO/IEC 27090:2026, these practice tests provide an effective way to assess your readiness and reinforce key concepts.

Unlike traditional training courses, this course is entirely focused on exam preparation and practical verification. Every question is designed to reflect real-world AI threat detection, risk mitigation, and secure engineering scenarios, helping you develop the analytical and decision-making skills expected of an AI Cybersecurity Specialist or Implementer.

What You'll Get

• High-quality practice questions

• 5 full-length practice exams, organized by certification domain

• Detailed explanations for every answer

• Scenario-based questions similar to those found in professional certification exams

• Questions covering both theoretical knowledge and practical AI security implementation

• Unlimited practice to reinforce learning and improve confidence

Practice Test Domains

Domain 1 – Applying Information Security & Governance to AI Systems Build a solid foundation in AI cybersecurity governance by mastering the core principles defined in Clauses 1 through 5 of ISO/IEC 27090:2026.

Topics covered include:

• Clauses 1–4 & Clause 5.1 – Scope, Normative Foundations & Integration:

  • Understanding the scope and terminology of ISO/IEC 27090:2026

  • Complementing established cybersecurity practices with AI-specific measures

  • Applying ISO/IEC 27002 controls where determined by risk assessments

  • Integration with ISO/IEC 27001 (ISMS) and ISO/IEC 42001 (AIMS)

• Clause 5.2 – AI Asset Inventory & Governance Programme:

  • Cataloging in-house models, hosted APIs, RAG knowledge bases, and AI agents

  • Identifying and protecting training datasets, fine-tuning pipelines, and model weights

  • Establishing ownership and accountability across organizations developing or using AI

• Clause 5.3 – Zero Trust Principles for AI:

  • Treating model inputs, external retrieval sources, and model outputs as untrusted

  • Enforcing least-privilege access across AI pipelines and components

• Clause 5.4 – Secure AI Software Engineering Practices:

  • Integrating security requirements across AI development, operation, and model updates

• Clause 5.5 – AI Supply Chain Risk Management:

  • Vetting third-party foundation model providers, open-source repositories, and pre-trained weights

  • Managing software dependencies, cryptographic signing, and behavioural verification (5.5.4, 5.5.5)

• Clauses 5.6 & 5.7 – AI Threat Modelling & Scoped Red Teaming:

  • Conducting structured AI threat modeling per system

  • Executing adversarial red-team testing prior to deployment and after major updates

Domain 2 – AI Model & Data Integrity Compromises (Poisoning & Evasion) This section focuses on the first part of Clause 6 (including Clause 6.12), addressing compromises that corrupt how an AI system learns, behaves, or classifies data.

Topics include:

• Data Poisoning:

  • Threat vectors across initial training, fine-tuning, RAG ingestion, and continuous learning

  • How corrupted operational inputs become future training data during retraining

  • Failure modes, bias injection, availability degradation, detection, and mitigation

• Direct Model Poisoning:

  • Tampering directly with model weights, architectures, serialization files, or pre-trained adapters

  • Identifying and preventing backdoor insertion in pre-trained models

• Evasion Attacks:

  • Adversarial input manipulation (imperceptible perturbations in vision, audio, or text)

  • Bypassing security classifiers and forcing misclassifications at inference time

  • Defensive hardening against adversarial examples

• Clause 6.12 – Data and System Provenance:

  • Establishing verifiable lineage and auditability for datasets and models

  • Cryptographic hashing and integrity tracking across training and update pipelines

• Limiting Data Exposure in Training:

  • Data sanitization and minimization prior to training or fine-tuning to reduce compromise impact

Domain 3 – AI Confidentiality, Exfiltration, Injection & Agent Controls This is the largest domain of the course and covers the second half of Clause 6 (Clauses 6.5 to 6.11), reflecting the critical emphasis placed on intellectual property protection, data privacy, LLM security, and autonomous AI agents.

Topics include:

• Model Exfiltration and Model Theft:

  • Protecting proprietary models from direct artifact theft

  • Detecting and mitigating systematic API querying (model extraction, cloning, and distillation)

• Membership Inference & Model Inversion:

  • Defending against attacks that probe output distributions or confidence scores to deduce whether specific records were in the training set (membership inference)

  • Preventing attackers from reconstructing private training data features (model inversion)

• Clause 6.8 – Sensitive Model Output:

  • Preventing unintended disclosure of PII, confidential business data, or proprietary system context in generated outputs

• Clause 6.9 – Prompt Injection (Direct & Indirect):

  • Defending against malicious user prompts designed to override system instructions (direct injection)

  • Mitigating indirect prompt injections embedded in external retrieval sources, web pages, emails, and RAG documents

• Clause 6.10 – Outputs Containing Injection Attacks:

  • Validating and sanitizing model outputs before passing them to downstream systems to prevent SQL injection, Cross-Site Scripting (XSS), or command execution

• Clause 6.11 – Controlling Unwanted Model Behaviour & AI Agent Security:

  • Distinguishing malicious attacks from benign model hallucinations (6.11.3.1)

  • Minimizing published technical details about internal model architectures

  • Securing autonomous AI agents (6.11.3.3) through separate agent identities, task-scoped permissions with expiration, and human-in-the-loop approval for high-impact actions

Domain 4 – Lifecycle Mitigation Dynamics, Monitoring & Runtime Defense Learn how to maintain, monitor, and adapt AI security safeguards throughout the operational lifecycle as defined in Clause 7 (Clauses 7.1 to 7.9) of ISO/IEC 27090:2026.

Topics include:

• Clause 7.2 – Conflicts Between Mitigations:

  • Analyzing operational trade-offs where defending against one threat (e.g., adversarial robustness training) increases vulnerability to another (e.g., membership inference or privacy leakage)

• Clauses 7.3 & 7.4 – Continuity Across the Lifecycle & Control Deterioration:

  • Maintaining security handoffs between development, deployment, and operations teams

  • Addressing how AI models and security controls degrade over time due to concept drift and evolving adversarial techniques

• Clause 7.5 – AI-Specific Logging and Monitoring:

  • Designing telemetry and audit trails that record user identity, timestamp, raw input, generated output, tool calls, and the exact model version for incident reconstruction

• Clause 7.6 – Protecting AI Development Environments:

  • Securing notebooks, feature stores, experiment trackers, and training clusters

• Clauses 7.7 & 7.8 – Detecting Malicious Inputs & Throttling Model Use:

  • Implementing runtime input guardrails and anomaly detection

  • Enforcing query rate-limiting and API throttling to slow down automated extraction and probing

• Clause 7.9 – Keeping Models Small (Principle of Minimality):

  • Reducing model size, parameter count, and capability scope to limit training data memorization and shrink the attack surface

Domain 5 – Asset-Lifecycle Threat Mapping, Conventional AI Twists & Federated Learning This domain covers Annexes A, B, and C of ISO/IEC 27090:2026, focusing on specialized threat scenarios, infrastructure-level AI compromises, and decentralized machine learning architectures.

Topics include:

• Annex A – Mapping Attacks to AI Lifecycle Stages & Assets:

  • Connecting specific threat vectors to targeted assets (training data, model parameters, inputs, outputs) across each stage of the AI lifecycle

• Annex B (B.2–B.7) – AI-Specific Versions of Conventional Attacks:

  • Sponge and Resource Exhaustion Denial-of-Service (DoS) attacks (B.2.1) engineered to maximize compute time and energy consumption

  • Manipulated Random Number Generators (RNG) skewing stochastic training and initialization processes

  • Backdoors embedded in serialized model files (such as malicious pickle payloads or architectural triggers)

  • Conventional infrastructure breaches resulting in the theft of model weights or leaks of training and input data

• Annex C – Federated Machine Learning Security:

  • Understanding decentralized training architectures where raw data remains on local nodes

  • Mitigating poisoned local gradient/weight updates and gradient inversion attacks in federated environments

All questions in this domain are scenario-based, requiring you to connect complex attack behaviors to their potential business and safety-critical impacts.

Why Take This Course?

This course is ideal for cybersecurity professionals, AI engineers, auditors, and certification candidates preparing for an ISO/IEC 27090:2026 exam, PECB AI security credentials, or ISO/IEC 42001/27001 technical assessments. Each practice test is designed to reinforce understanding, improve exam readiness, and build confidence by exposing learners to realistic exam-style questions.

By completing all five practice tests, you will strengthen your understanding of the ISO/IEC 27090:2026 standard, the AI threat and compromise catalogue, lifecycle defense trade-offs, and the practical skills required to protect training data, AI models, and intellectual property.

Disclaimer:

This course provides practice examinations for educational purposes. It is not affiliated with, endorsed by, or sponsored by ISO, PECB, CQI/IRCA, BSI, or any certification body. ISO/IEC 27090® is a registered standard, and learners should refer to the official standard and their chosen certification provider for authoritative requirements and exam information.