Are you preparing for the ISTQB Certified Tester – AI Testing (CT-AI) v2.0 certification and want to assess your readiness with realistic, high-quality exam‑style practice questions?

This comprehensive practice exam course has been designed to mirror the real CT-AI v2.0 certification exam as closely as possible, covering the updated syllabus released in April 2026 (Chapters 1–7)..

With 6 full‑length practice tests containing 240 questions in total, you will gain the confidence and knowledge required to pass the ISTQB CT-AI v2.0 certification on your very first attempt. Each question is carefully written to match the difficulty, structure, and exam‑style wording you will face on test day, including K1–K4 cognitive levels and hands‑on exercise scenarios (H1–H2).

Every question comes with detailed explanations for both correct and incorrect answers, ensuring that you not only know the right answer but also understand why the other options are wrong. This unique approach deepens your understanding and prepares you for any variation of the question that may appear in the real exam, covering topics such as ML workflows, neural network coverage measures, bias testing, metamorphic testing, drift detection, adversarial testing, and ML deployment testing.

Our ISTQB CT-AI v2.0 practice exams will help you identify your strong areas and pinpoint where you need improvement across all seven chapters. By completing these tests under timed conditions, you will build the exam discipline and confidence required to succeed.

This course is updated to stay 100% aligned with the latest ISTQB CT-AI v2.0 syllabus (2026 release).

Version Upgrade History:

<Updated all Details on 12-May-2026>

Comprehensive Coverage for CT-AI v2.0

This comprehensive practice exam course is designed to help AI testers, QA engineers, developers, and professionals assess readiness, reinforce concepts, and master the ISTQB CT-AI v2.0 certification.

Each mock test is carefully crafted to cover 100% of the official v2.0 syllabus, including: AI fundamentals (narrow/general/super AI, generative AI), ML workflows (supervised/unsupervised/reinforcement learning), neural networks and coverage measures (neuron, kMNC, NBC), bias testing (data and algorithmic bias), ISO/IEC 25059 quality characteristics (AI correctness, robustness, transparency, intervenability), safety considerations, test oracle challenges, red teaming for GenAI/LLM, input data testing (data pipeline, representativeness, label correctness, dataset constraints), model testing (metamorphic, adversarial, drift, overfitting/underfitting, A/B, back-to-back testing), and ML development/deployment testing (shadow, canary, rollback, API testing).

This course is regularly updated to stay 100% aligned with ISTQB CT-AI v2.0 and evolving AI testing practices.

Why This ISTQB CT-AI v2.0 Practice Exam Course is Unique

• 6 Full-Length Mock Exams: Total 240+ questions simulating the real ISTQB CT-AI v2.0 exam structure and chapter weightings.

• 100% Syllabus Coverage: Covers all K-level topics from K1 (Remember) to K4 (Analyze) as defined in the official v2.0 syllabus.

• Diverse Question Categories: This course ensures comprehensive preparation across all ISTQB CT-AI knowledge levels:

  • K1 – Remember: Recall key facts, definitions, and AI/ML terminology (e.g., types of ML, confusion matrix terms, coverage measures).

  • K2 – Understand: Explain and interpret AI testing concepts, ML workflows, quality characteristics from ISO/IEC 25059, and test oracle solutions.

  • K3 – Apply: Use AI testing principles and methods in practical scenarios (e.g., calculate ML functional performance metrics, apply metamorphic testing, implement red teaming).

  • K4 – Analyze: Break down complex AI systems to identify biases, model drift (data/concept), overfitting/underfitting, and adversarial vulnerabilities.

• Real Exam-Like Format: Multiple-choice and select-all-that-apply questions with balanced answer distribution.

• Comprehensive Explanations: Each question includes detailed rationales for all answer options, helping you learn why answers are correct or incorrect.

• Latest Syllabus Alignment (v2.0): Topics include generative AI, locked vs. adaptive AI systems, statistical testing (MoE, CL), red teaming, data pipeline testing, metamorphic testing, drift testing, A/B and back-to-back testing, and ML development/deployment testing.

• Every question is mapped to its relevant chapter (1–7) and learning objective, helping learners track syllabus coverage effectively.

• Scenario-Based Questions: Real-world, practical examples replicating ISTQB CT-AI v2.0 exam conditions, including hands‑on exercise scenarios (H1–H2).

• Exam Weightage Distribution: Questions follow official chapter weightings (Ch1:15%, Ch2:7.5%, Ch3:17.5%, Ch4:17.5%, Ch5:15%, Ch6:22.5%, Ch7:5%) for strategic preparation.

• Timed Practice: Simulate realistic exam durations for time management and confidence.

• Ideal for AI Testers & QA Engineers: Build skills for ISTQB v2.0 certification and real-world AI testing of ML systems.

• Randomized Question Bank: Questions and options reshuffle in each attempt to prevent memorization and encourage active learning.

• Performance Analytics: Receive chapter‑wise and domain‑wise insights to identify strengths and improvement areas, focusing preparation on topics like responsible AI, drift detection, adversarial testing, and model deployment.

• Practical, Real-World Application: Reinforce knowledge through scenario‑based and problem‑solving questions across all v2.0 syllabus topics, including input data testing and ML model testing.

Exam Details – ISTQB Certified Tester – AI Testing (CT-AI) v2.0

• Exam Body: ISTQB (International Software Testing Qualifications Board)

• Exam Name: ISTQB Certified Tester – AI Testing (CT-AI) v2.0

• Prerequisite Certification: ISTQB Certified Tester Foundation Level (CTFL) v4.0 or later

• Exam Format: Multiple Choice Questions (MCQs) – single answer and select‑all‑that‑apply questions

• Certification Validity: Lifetime (no renewal required)

• Number of Questions: 40

• Total Points: 44 points (combination of 1‑point and 2‑point questions based on cognitive level K1‑K2 vs K3‑K4)

• Passing Score: 29 points out of 44 points (approximately 66%)

• Exam Duration: 60 minutes (75 minutes for candidates whose native language is not the exam language)

• Question Weightage: Varies – knowledge questions (K1‑K2) are 1 point each; application/analysis questions (K3‑K4) are 2 points each

• Language: English (localized versions may be available from ISTQB member boards)

• Syllabus Version: CT-AI v2.0 (released April 17, 2026)

• Exam Sections: All seven chapters are examinable (Introduction to AI, Quality Characteristics, Machine Learning, Testing AI‑Based Systems, Input Data Testing, Model Testing, ML Development Testing) following the official weightings

• Hands‑on Objectives (H1‑H2): Not directly examinable but support understanding of K3‑K4 scenario‑based questions

  
  

Detailed Syllabus and Topic Weightage

The ISTQB CT-AI 2.0 certification exam evaluates your understanding of AI testing principles, machine learning testing, quality characteristics, AI test automation, and practical application of testing AI-based systems. The syllabus is divided into 7 Domains, covering knowledge levels K1–K4, with question distribution reflecting topic weightage.

Domain 1: Introduction to Artificial Intelligence (~15% – 6 questions)

• AI definitions (narrow AI, general AI, super AI)

• AI-based systems vs. conventional systems

• Generative AI (GenAI) and frontier AI

• AI technologies: ML, deep learning (CNN, RNN, transformers), NLP, computer vision

• Hardware for ML: CPU, GPU, ASIC, neuromorphic processors

• AI model development and hosting: on-premises, cloud, AI as a Service (AlaaS)

• ML development frameworks (e.g., TensorFlow, PyTorch)

• Regulations and standards: EU AI Act, OECD AI Principles, ISO/IEC standards

Domain 2: Quality Characteristics for AI-Based Systems (~7.5% – 3 questions)

• ISO/IEC 25059 quality model extensions

• AI functional correctness (threshold-based acceptance)

• Functional adaptability, user controllability, transparency

• AI robustness, intervenability

• Societal and ethical risk mitigation (fairness, accountability, sustainability)

• Safety considerations in AI-based systems (non-determinism, self-learning, explainability)

• Acceptance criteria for AI-based systems (statistical, probabilistic, threshold-based)

Domain 3: Machine Learning (~17.5% – 7 questions)

• Supervised learning (classification, ML regression)

• Unsupervised learning (clustering, association)

• Reinforcement learning

• ML workflow: objectives, framework selection, algorithm selection, data preparation, training, evaluation, tuning, testing, deployment, monitoring

• Pretrained models, fine-tuning, Retrieval-Augmented Generation (RAG)

• Training, validation, and test datasets (including k-fold cross-validation)

• Neural network structure: input/hidden/output layers, neurons, activation functions, weights, biases

• Coverage measures: neuron coverage, k-multisection neuron coverage (kMNC), neuron boundary coverage (NBC)

Domain 4: Input Data Testing for Machine Learning Systems (~15% – 6 questions)

• Data preparation activities: acquisition, preprocessing (cleaning, transformation, augmentation), feature engineering

• Exploratory Data Analysis (EDA)

• Input data risks and mitigations (bias, poisoned data, missing data, imbalance)

• Testing for bias: data bias, algorithmic bias, disparate impact analysis, counterfactuals

• Data pipeline testing: component, integration, system, system integration, production testing

• Data representativeness testing: target population definition, statistical assessment (Chi-squared, Kolmogorov-Smirnov)

• Dataset constraint testing: single-value constraints (missing, range, type), multi-value constraints (sum, count, duplicate, outlier), comparison constraints (greater than, correlate)

• Label correctness testing: expert review, multiple annotation (IAA, Cohen’s Kappa), model loss analysis, confidence score analysis

Domain 5: Model Testing for Machine Learning Systems (~22.5% – 9 questions)

• ML model risks and mitigations (bias, overfitting, underfitting, adversarial examples, drift)

• ML model documentation and review (Model Cards, Datasheets for Datasets, transparency)

• ML functional performance testing for probabilistic systems (margin of error - MoE, confidence level - CL, sample size calculation)

• Adversarial testing: adversarial examples, black-box vs. white-box, transferability

• Metamorphic testing: metamorphic relations (MRs), source/follow-up test cases, monotonicity, invariance, consistency

• Drift testing: data drift, concept drift, static vs. dynamic drift detection

• Overfitting and underfitting detection (test dataset evaluation, learning curves)

• A/B testing for model comparison

• Back-to-back testing (pseudo-oracles, independent development)

Domain 6: Testing AI-Based Systems Overview (~17.5% – 7 questions)

• Locked vs. adaptive AI-based systems (testability implications)

• Rationale for statistical approach (non-determinism, distributional performance, uncertainty, bias, regulatory context)

• Test oracle problem: probabilistic nature, incomplete specifications, complexity, subjectivity, self-learning

• Solutions: output/environmental boundaries, expert consultation, specialized testing (A/B, back-to-back, metamorphic), proxy oracles

• Testing Generative AI and LLMs: black-box testing, input explosion problem, context window, benchmark suites, red teaming

• Red teaming: fault attacks, harmful capabilities (security, safety, fairness, bias, misinformation)

• Exploratory testing of LLMs

• Test levels for ML systems: input data testing, model testing, component, integration, system, acceptance

• Risk-based testing for ML systems (project risks: development, framework; product risks: input data, model)

Domain 7: Machine Learning Development Testing (~5% – 2 questions)

• ML development risks and mitigations: API defects, framework selection/installation, algorithm/hyperparameter selection, data allocation, evaluation approach, security vulnerabilities

• Test approaches: API testing, framework suitability review, bias testing, smoke testing, performance testing, usability testing, security testing, A/B testing, back-to-back testing

• Deployment testing types:

  • Installability testing

  • Rollback testing

  • Canary testing

  • Shadow testing

  • Model conversion testing

  • Cross-device testing

  • API testing

Practice Test Structure & Preparation Strategy

• Prepare for the ISTQB Certified Tester – AI Testing (CT-AI) v2.0 certification exam with realistic, exam-style mock tests that build conceptual understanding, hands-on readiness, and exam confidence.

• 6 full-length practice tests: 6 complete mock exams with 40 questions each (240 total questions), timed and scored, reflecting the real exam structure, style, and complexity of the updated v2.0 syllabus.

• Diverse question categories: Questions are designed across multiple cognitive levels (K1–K4) as per the latest CT-AI v2.0 learning objectives.

• Knowledge-heavy questions (K1–K2): Worth 1 point each, focus on recalling theory, definitions, and basic AI/ML concepts (~50% of questions).

• Application & analysis questions (K3–K4): Scenario-based or analytical, worth 2 points each, testing application, reasoning, and analysis (~50% of total points).

• Hands-on elements (H1–H2): Practical activities from Chapters 3, 4, 5, and 6 (Machine Learning, Testing AI-Based Systems, Input Data Testing, Model Testing) to reinforce real-world AI testing tasks.

• Comprehensive explanations: Detailed reasoning for correct and incorrect options to enhance learning and address common misconceptions.

• Timed & scored simulation: Practice under realistic exam timing to develop focus, pacing, and endurance for the actual certification exam.

• Randomized question bank: Questions and answer options reshuffle in each attempt to prevent memorization and ensure genuine understanding.

• Performance analytics: Domain-wise insights to identify strengths and areas for improvement, with focus on ML functional performance metrics, bias detection, test oracles, metamorphic testing, and drift testing.

Question 1

Which term describes the phenomenon where a large language model produces output that is grammatically fluent and contextually plausible but factually incorrect?

Options:

A. Forgetting
B. Hallucination
C. Overfitting
D. Underfitting

Answer: B

Explanation:

• A: Incorrect. Forgetting (or catastrophic forgetting) describes a model losing previously learned knowledge when trained on new tasks or data. This is a training dynamics issue, not the term for generating plausible but factually incorrect output during inference.

• B: Correct. Hallucination is the established term for this phenomenon, where an LLM generates output that is grammatically fluent and contextually plausible but contains factually incorrect or fabricated information not grounded in the input or verifiable reality, as covered in the syllabus.

• C: Incorrect. Overfitting occurs when a model learns training data too specifically, including noise, and performs poorly on unseen data. It describes a generalization failure during training, not inference‑time content inaccuracy.

• D: Incorrect. Underfitting happens when a model fails to learn sufficient patterns from training data. It is a learning process failure, not a term for producing fluent but factually wrong outputs.

Domain: Machine Learning
K-Level: K1 – Remember

Question 2

A test manager at a software house is evaluating three candidate LLM tools for integration into the test automation pipeline. One criterion is the extent to which each tool supports subsequent adaptation to the organization's specific test domain without retraining from scratch. Given the following evaluation criteria:

a. The tool supports fine-tuning on custom datasets to adapt outputs to domain‑specific terminology and test patterns.

b. The tool provides pre‑built connectors for existing test management platforms.

c. The organization can export and own the adapted model weights after fine‑tuning.

d. The vendor guarantees inference latency below 200 ms for 99% of requests.

Which combination of criteria BEST reflects fine‑tuning potential as a selection consideration?

Options:

A. a and c
B. c and d
C. a and b
D. b and d

Answer: A

Explanation:

• A: Correct. Criterion a (support for fine‑tuning on custom datasets) directly addresses the ability to adapt the model to domain‑specific needs. Criterion c (ownership of adapted model weights) ensures the organization retains control without vendor lock‑in. Together, they fully capture fine‑tuning potential.

• B: Incorrect. Criterion d (inference latency) is a performance SLA metric unrelated to fine‑tuning capability. While criterion c is relevant, pairing it with an operational metric misses the requirement for training‑based adaptation.

• C: Incorrect. Criterion b (pre‑built connectors) reflects integration compatibility, not fine‑tuning potential. Criterion a is correct, but the combination fails to include ownership of model weights (criterion c), which is essential for assessing domain adaptation without dependency.

• D: Incorrect. Both criteria b (connectors) and d (latency) describe operational and interoperability characteristics. Neither measures the tool’s capacity to support fine‑tuning on custom datasets for domain adaptation.

Domain: Input Data Testing for ML Systems
K-Level: K2 – Understand

Question 3

A test lead at a financial platform organisation is integrating an LLM‑based defect classification assistant into the regression testing workflow for a core banking transaction processing system. Following initial deployment, the lead observes that the assistant occasionally classifies defects as resolved when the actual defect is still present and replicable in the test environment. To address this hallucination risk, the lead must implement consistency checks that detect when the LLM output contradicts verifiable evidence available in the test environment. Considering the operational constraints of a live regression workflow, which of the following consistency check strategies MOST effectively detects LLM hallucination in defect classification outputs for this scenario?

Options:

A. Implement an automated cross‑validation step that re‑executes the test case associated with each LLM‑classified defect and compares the live execution result against the LLM classification before accepting the output as valid.
B. Implement a keyword frequency filter that rejects LLM outputs containing high‑frequency tokens associated with resolution claims, substituting a default unresolved classification for all flagged outputs.
C. Implement a human review gate that requires a test engineer to manually approve every LLM defect classification output before it is committed, irrespective of confidence score.
D. Implement a cosine similarity threshold filter that discards LLM outputs falling below a defined semantic similarity score relative to the defect description, replacing them with the most similar historical classification.

Answer: A

Explanation:

• A: Correct. Automated cross‑validation re‑executes the test case and compares the live result with the LLM classification. This directly detects hallucinated resolution claims by providing ground‑truth execution evidence, making it the most reliable and scalable strategy for a live regression workflow.

• B: Incorrect. A keyword frequency filter relies on lexical pattern matching, not factual verification. Hallucinated resolution claims and correct resolutions both contain similar keywords, so this filter cannot distinguish between them and will produce false positives/negatives.

• C: Incorrect. Manual review of every output is not scalable in a high‑volume continuous integration pipeline. It also does not leverage verifiable test environment evidence; it substitutes human judgement for automated consistency checking, which is inefficient and error‑prone.

• D: Incorrect. Cosine similarity measures semantic relatedness, not factual correctness. A hallucinated claim can be semantically similar to the defect description while being false. Replacing outputs with historical classifications does not verify current defect state.

Domain: Machine Learning
K-Level: K3 – Apply

Preparation Strategy & Study Guidance

• Understand the concepts, not just the questions: Use these tests to identify weak areas, but supplement study with the official ISTQB CT-AI v2.0 syllabus (Chapters 1–7) covering AI fundamentals, generative AI, locked/adaptive systems, input data testing, model testing, and ML development testing

• Target 80%+ in practice tests: The real exam requires 29 out of 44 points to pass (≈66%); achieving higher scores in practice builds confidence and mastery across all cognitive levels K1–K4

• Review explanations in detail: Carefully study why each answer is correct or incorrect to avoid conceptual mistakes, especially for scenario‑based questions on bias detection, metamorphic testing, drift detection, and adversarial testing

• Simulate real exam conditions: Attempt mock tests in timed, distraction‑free sessions to develop focus, speed, and exam endurance (60 minutes for 40 questions)

• Hands‑on application: Reinforce AI testing knowledge through practical examples aligned with the syllabus hands‑on objectives (H1–H2) such as creating an ML model, performing data preparation, evaluating with ML functional performance metrics, applying metamorphic testing, and exploratory testing of an LLM

Why This Course Is Valuable

• Realistic exam simulation aligned with ISTQB CT-AI v2.0 format including knowledge levels K1 (Remember), K2 (Understand), K3 (Apply), and K4 (Analyze)

• Full syllabus coverage including AI fundamentals (narrow/general/super AI, generative AI), ML workflows (supervised/unsupervised/reinforcement), neural networks and coverage measures, ISO/IEC 25059 quality characteristics, bias testing, data pipeline testing, representativeness, label correctness, model testing (metamorphic, adversarial, drift, overfitting/underfitting, A/B, back‑to‑back), and ML development/deployment testing (shadow, canary, rollback, API testing)

• In‑depth explanations for correct and incorrect answers to improve understanding of test oracles, statistical testing (margin of error, confidence level), and red teaming for GenAI/LLM

• Timed, scored tests with randomized questions for better preparation following official chapter weightings (Ch1:15%, Ch2:7.5%, Ch3:17.5%, Ch4:17.5%, Ch5:15%, Ch6:22.5%, Ch7:5%)

• Designed for AI testers, QA engineers, and developers preparing for ISTQB CT-AI v2.0

• Updated as per the latest ISTQB CT-AI v2.0 syllabus (April 2026 release)

Top Reasons to Take This Practice Exam

• 6 full‑length mock exams with 240+ questions

• 100% coverage of official ISTQB CT-AI v2.0 syllabus

• Realistic multiple‑choice and select‑all‑that‑apply questions

• Detailed rationales for correct and incorrect answers

• Balanced question distribution across K1–K4 levels

• Timed simulations to replicate exam conditions

• Randomized question bank for active learning

• Accessible anywhere, anytime on desktop or mobile

• Lifetime updates included for syllabus changes

What This Course Includes

• 6 Full‑Length Practice Tests: Simulate real exam conditions to test your readiness, each with 40 questions (44 total points)

• Access on Mobile: Study anytime, anywhere on your phone or tablet

• Full Lifetime Access: Learn at your own pace with no expiration

• Money‑Back Guarantee: 30‑day no‑questions‑asked refund policy

Who This Course Is For

• Professionals preparing for the ISTQB CT-AI v2.0 exam

• QA engineers, test leads, and automation testers entering AI testing for ML systems

• Developers and IT professionals enhancing AI testing skills (input data testing, model testing, deployment testing)

• AI/ML enthusiasts aiming for ISTQB AI Testing Certification v2.0

• Professionals addressing real‑world AI testing challenges like bias, data drift, concept drift, adversarial vulnerabilities, non‑deterministic outputs, and test oracle problems

• Career changers seeking expertise in AI QA and test automation for ML workflows

What You’ll Learn

• Core AI and ML principles, including neural networks, coverage measures (neuron coverage, kMNC, NBC), and ML workflows (training, validation, testing, deployment, monitoring)

• AI test design, execution, and validation techniques specific to locked vs. adaptive AI‑based systems

• Bias detection (data bias, algorithmic bias), explainability (XAI), transparency, AI robustness, and AI system safety according to ISO/IEC 25059

• Scenario‑based testing for ML systems using metamorphic testing, adversarial testing, drift testing, A/B testing, and back‑to‑back testing

• Input data testing techniques: data pipeline testing, representativeness testing, dataset constraint testing, label correctness testing

• ML development and deployment testing: shadow testing, canary testing, rollback testing, model conversion testing, API testing

• Time management and exam strategies for ISTQB CT-AI v2.0

• Practical knowledge to confidently pass the ISTQB CT-AI v2.0 certification exam

Requirements / Prerequisites

• Prior ISTQB Foundation Level (CTFL) certification required (v4.0 or later)

• Basic understanding of software testing principles

• Familiarity with AI, ML, or neural network concepts is helpful but not required

• Computer with internet access for online mock exams

• Curiosity to learn AI testing, bias detection, drift monitoring, and AI system lifecycle testing