Study guide

1.1 Current Exam Facts (CT-AI v2.0)

ISTQB Certified Tester AI Testing CT-AI

Track your progress as you read

Finished sections get a check mark.

1 Introduction & Exam Overview

1.1 Current Exam Facts (CT-AI v2.0) 1.2 Scoring, K-levels & Chapter Weights

2 Introduction to AI: Systems and Capabilities

2.1 AI-Based vs Conventional Systems 2.2 Narrow AI, General AI and Super AI 2.3 Types of AI Technologies 2.4 Generative AI

3 AI Hardware, Hosting, Frameworks and Governance

3.1 Hardware for Machine Learning Systems 3.2 Development and Hosting of AI Models 3.3 Machine Learning Development Frameworks 3.4 Regulations and Standards for AI

4 Quality Characteristics and Acceptance Criteria

4.1 ISO/IEC 25059 AI Quality Characteristics 4.2 AI and Safety 4.3 Acceptance Criteria for AI-Based Systems

5 Machine Learning Forms, Workflow and Data Preparation

5.1 Forms of Machine Learning 5.2 The Machine Learning Workflow 5.3 Pretrained Models, Fine-Tuning and RAG 5.4 Data Preparation Activities

6 Datasets, Classification Metrics and Neural Networks

6.1 Training, Validation and Test Datasets 6.2 Calculating Classification Metrics 6.3 Structure of Deep Neural Networks 6.4 Neural Network Coverage Measures

7 Testing Challenges for AI-Based Systems

7.1 Locked vs Adaptive AI-Based Systems 7.2 Why Statistical Testing Is Needed 7.3 Test Oracles for AI-Based Systems

8 GenAI Testing, Test Levels and Risk-Based Testing

8.1 Testing Generative AI 8.2 Red Teaming for GenAI 8.3 Test Levels for Machine Learning Systems 8.4 Risk-Based Testing of ML Systems

9 Input Data Risks, Bias and Pipelines

9.1 Input Data Risks and Mitigations 9.2 Testing for Bias 9.3 Data Pipeline Testing

10 Representativeness, Constraints and Labels

10.1 Testing for Data Representativeness 10.2 Dataset Constraint Testing 10.3 Label Correctness Testing

11 Model Risks, Documentation and Performance Testing

11.1 ML Model Risks and Mitigations 11.2 ML Model Documentation and Review 11.3 Functional Performance Testing of Probabilistic MLS

12 Adversarial, Metamorphic and Drift Testing

12.1 Adversarial Testing 12.2 Metamorphic Testing 12.3 Drift Testing

13 Overfitting, A/B Testing and Back-to-Back Testing

13.1 Testing for Overfitting and Underfitting 13.2 A/B Testing for ML Systems 13.3 Back-to-Back Testing

14 ML Development and Deployment Testing

14.1 ML Development Risks and Mitigations 14.2 ML System Deployment Testing

1.1 Current Exam Facts (CT-AI v2.0)

Key Takeaways Explain This Topic AI

ISTQB Certified Tester AI Testing Version 2.0 is the current syllabus; general availability was 17 April 2026, and it replaces Version 1.0.

English Version 1.0 exams and training remain available until 21 April 2027; other languages remain available until 21 October 2027.

The live exam is 40 multiple-choice questions in 60 minutes, or 75 minutes if the exam language is not the candidate's native language (+25% extra time).

The pass mark is 65%, implemented as 29 of 44 points because questions are weighted; 26 of 40 items is not the official passing score.

ISTQB does not publish a global exam fee; national boards and exam providers set prices, and U.S. sittings often run through ASTQB and providers such as AT*SQA.

Last updated: September 2026

What CT-AI Version 2.0 is

The International Software Testing Qualifications Board (ISTQB) Certified Tester AI Testing (CT-AI) credential is a specialist exam for people who test artificial intelligence (AI) -based systems. Version 2.0 is the current syllabus. ISTQB released it for general availability (GA) on 17 April 2026 . From that date, Version 2.0 is the syllabus this guide treats as current, and it replaces Version 1.0 as the up-to-date examination baseline.

This ExamFinch chapter is independent teaching. ExamFinch is not an ISTQB member board, accredited training provider, or exam provider, and this guide is not an official ISTQB publication. Use it to understand the published rules and to practice the ideas those rules reward. Download the official syllabus, sample exam, and Exam Structures and Rules tables from ISTQB when you want the primary source.

A useful analogy is a specialist endorsement on top of a basic professional license. The required entry ticket is the Certified Tester Foundation Level (CTFL) certificate. CTFL proves you already know core testing vocabulary: test cases, oracles, coverage, and defect reporting. CT-AI then asks whether you can apply that craft to systems that learn from data, produce probabilistic outputs, and often refuse to give a single expected result for every input.

What Version 2.0 actually tests

Version 2.0 focuses on testing AI-based systems , including machine learning (ML) systems and generative AI (GenAI) such as large language models (LLMs) . Typical product pictures help:

A hospital trains an image classifier to flag likely fractures on X-rays. Your job is not to invent a better neural network. Your job is to test the data that fed training, the model that came out, the surrounding clinical workflow, and the risks that appear after go-live.

A retailer ships a customer-support LLM that answers return-policy questions. Your job includes exploratory testing and red teaming : can a user coax the model into inventing refunds, leaking internal prompts, or ignoring safety filters?

A bank scores loan applications with a locked ML model. Your job includes checking whether the training labels were representative, whether performance metrics match the business risk, and whether the model still behaves after the applicant population shifts.

Those examples share a theme. The system under test is not a conventional if-then program whose every branch you can trace with a debugger. It is a data-dependent, often non-deterministic product. CT-AI Version 2.0 is the ISTQB specialist exam written for that product class.

The split with CT-GenAI

Version 1.0 mixed two different skills: testing AI-based products, and using AI to support testing (for example, asking a model to draft test cases for a payroll system). Version 2.0 removed the second skill from CT-AI. ISTQB now places that subject with Certified Tester Testing with Generative AI (CT-GenAI) .

Keep the split straight on exam day. If the scenario is 'how do we test the chatbot we are shipping?', you are in CT-AI Version 2.0. If the scenario is 'how do we use a chatbot to help testers write cases for a non-AI billing engine?', that is CT-GenAI, not this exam. Candidates who still need both skills sit two different specialist exams; there is no combined shortcut hiding inside Version 2.0.

Version 1.0 sunset, certificates, and the missing bridge

ISTQB is not switching the lights off overnight. English-language Version 1.0 training and examinations, including retakes, remain available until 21 April 2027 . Other languages remain available until 21 October 2027 . After those dates, Version 1.0 exams and training courses are withdrawn.

If you already hold a Version 1.0 certificate, it remains valid . ISTQB does not require you to sit Version 2.0 to keep the credential. Specialist certificates in this scheme are treated as lifetime credentials: they do not expire on a renewal calendar. Wanting the Version 2.0 certificate is a separate career choice, not a repair job on a lapsed card.

There is no bridging exam from Version 1.0 to Version 2.0. Anyone who wants the Version 2.0 certificate sits the Version 2.0 paper as a full, separate exam. That matters if a job posting asks specifically for Version 2.0 knowledge of input-data testing, model testing, or LLM red teaming. A Version 1.0 holder is still certified; they are not automatically credited with Version 2.0 coverage.

Choosing a version if you have not sat yet

If you are already deep into Version 1.0 notes and you will sit inside the English or translated window, finishing Version 1.0 can be the efficient path. If you are starting now, this ExamFinch guide covers Version 2.0 , because that is the current syllabus and the one organized around the ML lifecycle (input data, model, and ML development testing). Version 2.0 also added dedicated coverage of testing GenAI and LLMs, which Version 1.0 did not treat as a first-class testing problem in the same way.

Exam format you must memorize

The Version 2.0 exam format is published by ISTQB as follows:

  • Rule Version 2.0 figure
  • Question style 40 multiple-choice questions
  • Total points 44 (questions are weighted)
  • Pass mark 65%, implemented as 29 of 44 points
  • Standard duration 60 minutes
  • Extra time +25% when the exam is not in the candidate's native language ( 75 minutes )
  • Prerequisite ISTQB Certified Tester Foundation Level (CTFL)
  • Prior AI expertise Not required beyond CTFL; testing or development familiarity is recommended
  • The 26-versus-29 trap

Many candidates multiply 0.65 by 40 and write 26 on a flashcard. That arithmetic would be correct only if every question were worth one point. Four questions are K3 (Apply) items worth two points each, so the paper totals 44 points . Sixty-five percent of 44 is 28.6, and the official passing score is 29 points . Scoring 28 of 44 is 63.6% , which is a fail.

Work a few sitting patterns so the number becomes muscle memory:

32 correct one-point items and zero two-point items = 32 points (pass), even though you missed every application question.

25 one-point items and all four two-point items = 25 + 8 = 33 points (pass).

28 one-point items and zero two-point items = 28 points (fail). You answered most questions and still sat one point short.

27 one-point items plus one two-point item = 29 points (pass on the line).

The exam is therefore not a head-count of ticks. It is a point exam. Section 1.2 of this chapter maps which chapters carry the two-point items.

Time on the clock

Forty items in 60 minutes is 90 seconds per question. That is enough for a K2 'which statement is true?' item if you already know the distinction. It is tight for a K3 calculation or a red-teaming scenario you have never practiced. The +25% extra time rule is a published ISTQB provision for candidates sitting in a non-native language . It turns 60 minutes into 75 minutes . It is not a medical-accommodation process you invent on the morning of the exam, and it is not granted merely because CTFL was taken years ago. Confirm with your exam provider how they record native-language status before you book.

Prerequisite, audience, training, and fees

CTFL is mandatory. ISTQB does not offer a Version 2.0 sitting to candidates who lack Foundation Level. You do not need a prior AI research degree. The syllabus introduces AI and ML concepts before it asks you to test them. ISTQB still recommends some software testing or development background—for example several months as a tester, developer, data scientist, or user acceptance tester—because exam items assume you already know what a test level and a product risk are.

Who typically sits

ISTQB describes the audience as people involved in testing AI-based systems, including:

  • Testers, test analysts, and test engineers
  • Test managers and test consultants
  • Data analysts and data scientists
  • Software developers who build AI-based systems
  • User acceptance testers

It is also taken by people who need a structured picture rather than a hands-on testing role: project managers, quality managers, software development managers, business analysts, IT directors, and management consultants. If your job is 'I accept a vendor model' or 'I own quality for a team that ships an LLM', you are in the intended audience even if you never write Python.

Training hours versus self-study

For accredited courses, the Version 2.0 syllabus requires a minimum of 19.5 hours of instruction, typically delivered as a three-day class. Accredited training is recommended because a member board has reviewed the courseware, and some providers include the exam sitting in the course. Self-study is allowed. You may prepare from the official syllabus, the sample exam, and independent materials such as this ExamFinch guide.

Accreditation of a training provider is a member-board process. It is not the same thing as ExamFinch claiming any ISTQB status. If you want an accredited classroom, search ISTQB's training-provider directory. If you want independent written teaching, you are already in the right place.

Fees are local, not global

ISTQB does not publish a single worldwide price. National boards and exam providers set fees. In the United States , delivery often runs through the American Software Testing Qualifications Board (ASTQB) ecosystem and exam providers such as AT*SQA . Other countries use their own member boards and providers. Forum screenshots, a colleague's 2024 invoice, or a training-bundle quote from another region are not a global rate. Ask the provider that will invigilate your sitting for the current fee, retake policy, and whether extra time is already in the booking form.

How to use this independent guide

Treat ISTQB downloads as the rules of the game. Treat this chapter as teaching commentary with examples. When a number is examinable as a fact—40 questions, 44 points, 29 to pass, 60 or 75 minutes, 19.5 accredited hours, 17 April 2026 GA, 21 April / 21 October 2027 sunsets—this chapter repeats the published figure so you can quote it under time pressure. What this chapter will not do is invent a pass rate, a worldwide price, or a claim that ExamFinch is approved by ISTQB.

Loading diagram... CT-AI v2.0 scope versus CT-GenAI

CT-AI v2.0 exam duration