CT-AI v2.0: AI Testing

Is CT-AI the right certification for you?

CT-AI is for testers who need to test systems that contain AI: machine learning models, the data they learn from, and generative AI features such as chatbots built on large language models. It is about testing AI, not about using AI to test. That second topic belongs to CT-GenAI.

Choose it if:

• You test, or will soon test, a product with an ML model or an LLM feature inside it.
• You need to talk with data scientists about training data, bias, model metrics and drift.
• You want structured approaches such as metamorphic testing, adversarial testing and red teaming.
• You hold CTFL. The syllabus recommends at least six months of experience in development, testing, data science or user acceptance testing.

Skip it or wait if:

• You mainly want to use ChatGPT-style tools to write tests faster. CT-GenAI covers that.
• You have no exposure to AI-based products and no plan to work on one. The material sticks better when you can relate it to a real system.

Exam at a glance

These figures describe CT-AI v2.0. Use the listed point threshold, rather than assuming a fixed number of correct answers on a weighted exam.

Source: Official exam structures and rules. Appointment arrangements and any approved extra time must be confirmed with your provider.

How the exam is weighted

ISTQB publishes the question split per chapter in its exam structure tables, not on the certification page. Here it is next to the training time the syllabus sets for accredited courses.

Syllabus chapter	Exam questions	Share of exam	Course training time
1. Introduction to artificial intelligence	6	15%	120 min
2. Quality characteristics for AI-based systems	3	7.5%	45 min
3. Machine learning	7	17.5%	375 min
4. Testing AI-based systems	7	17.5%	195 min
5. Input data testing for ML systems	6	15%	180 min
6. Model testing for ML systems	9	22.5%	225 min
7. ML development testing	2	5%	30 min
Total	40	100%	1,170 min (19.5 hours)

What the numbers tell you: training time and exam weight do not line up. Chapter 3 gets the most training time (375 minutes, about a third of the course) but only 7 questions. Chapter 6, model testing, carries the most questions (9). Courses spend time on ML basics so you can follow the testing chapters, but the exam rewards the testing chapters.

Almost the whole exam tests understanding: 36 questions are K2 (1 point each) and only 4 are K3 (2 points each). You need 29 of 44 points to pass.

Source: ISTQB Exam Structures & Rules Tables v1.19 (August 19, 2026) and the CT-AI v2.0 syllabus.

Entry requirements

You need CTFL. Ask which certificate evidence your provider accepts before you buy or schedule the exam.

Also known as: CT AI, AI Testing, Certified Tester AI Testing.

Source: Official syllabus and qualification overview.

What the syllabus covers

• Chapter 1, introduction to AI: narrow, general and super AI, ML technologies, generative AI, AI hardware and frameworks, and regulations.
• Chapter 2, quality characteristics: the ISO/IEC 25059 quality model for AI systems, safety, and acceptance criteria.
• Chapter 3, machine learning: forms of ML, the ML workflow, data preparation, performance metrics, neural networks and coverage measures.
• Chapter 4, testing AI-based systems: locked and adaptive systems, statistical testing, test oracles, testing generative AI and LLMs, red teaming and test levels.
• Chapter 5, input data testing: data risks, bias testing, data pipelines, representativeness and label correctness.
• Chapter 6, model testing: functional performance testing, adversarial and metamorphic testing, drift, overfitting, A/B testing and back-to-back testing.
• Chapter 7, ML development testing: development risks and deployment testing.

Accredited courses include hands-on exercises, such as building a simple ML model, evaluating performance metrics, implementing a perceptron and exploratory testing of an LLM. The exam itself is multiple choice and does not ask you to write code, but it does expect you to work with ML performance metrics.

What changed in v2.0

ISTQB announced CT-AI v2.0 on April 21, 2026. According to the announcement and the syllabus:

• The scope now covers only testing AI-based systems. Content on using AI for testing moved out of CT-AI.
• The learning path is reorganized around data, ML models and system-level testing.
• Generative AI and large language models get dedicated coverage, including red teaming.
• Accredited training is shorter: three days instead of four.
• Hands-on exercises are built into the syllabus.

“Version 2.0 provides a clearer and more practical foundation for testers working with machine learning and generative AI technologies,” said Klaudia Dussa-Zieger, President of ISTQB® and Chair of the CT-AI Taskforce, in the release announcement.

Read the official CT-AI v2.0 announcement.

How to prepare

The suggestions below are our editorial advice, not ISTQB requirements.

1. Work small numerical examples before relying on a formula or a dashboard label.
2. Separate a data problem, a model problem and a problem integrating the model into a product.
3. Write acceptance criteria that can be checked repeatedly, while stating uncertainty and assumptions.
4. Practice identifying what additional evidence a scenario needs before declaring an AI system acceptable.

A five-week study plan

This is our suggested plan for a tester with about five hours a week. It is editorial guidance, not an ISTQB requirement.

Week	Focus	Output to aim for
1	Chapters 1 and 2	A glossary of AI terms in your own words and the ISO/IEC 25059 characteristics mapped to a product you know
2	Chapter 3	A worked confusion matrix with accuracy, precision, recall and F1 calculated by hand
3	Chapters 4 and 5	A test approach for an LLM chatbot and a data risk checklist for an ML feature
4	Chapters 6 and 7	Three metamorphic relations and one adversarial test idea for a model you know
5	Sample exam and repair	One timed sample and a second pass on chapters 4 and 6

Use the official sample exams

Take one sample under timed conditions before you open the answers, then review every guess as well as every wrong answer. Our sample exam guide shows how to turn a practice score into a study plan.

• ISTQB CT-AI Sample Exam Questions v2.2
• ISTQB CT-AI Sample Exam Answers v2.2

Moving from AI Testing v1.0

CT-AI v2.0 replaces v1.0. The official page says English v1.0 exams and training remain available until April 21, 2027. Non-English availability ends on October 21, 2027.

The retiring v1.0 exam lists 40 questions, 47 points and a 31-point pass threshold. The current v2.0 exam lists 40 questions, 44 points and a 29-point threshold. Both list 60 minutes. Check version labels on your syllabus, samples and booking.

Build a new learning-objective checklist for v2.0 instead of treating older notes as a complete map. Before you pay, confirm the exam version and availability with your provider.

Official retiring v1.0 page, Syllabus changes and certificate validity

Before you book

Before you pay for CT-AI, confirm the exam version, the language and your eligibility with the provider. Our booking checklist, country guides and exam day checklist walk you through it.

Common questions

How many questions are on the CT-AI v2.0 exam, and what is the pass mark?

The exam has 40 multiple-choice questions worth 44 points. You need 29 points (65%) to pass. The standard time is 60 minutes, with 25% extra time available on request for non-native speakers.

Do I need coding or math skills for CT-AI?

You do not write code in the exam, which is multiple choice. You should be comfortable with basic ML performance metrics, because the syllabus expects you to calculate and interpret them. Accredited courses add hands-on exercises.

What is the difference between CT-AI and CT-GenAI?

CT-AI teaches you to test AI-based systems, such as ML models and LLM features. CT-GenAI teaches you to use generative AI tools in your own testing work. Since v2.0, CT-AI no longer covers using AI for testing.

Which CT-AI chapter matters most?

Chapter 6, model testing for ML systems, with 9 of 40 questions. Chapters 3 and 4 follow with 7 questions each.

Can I still take CT-AI v1.0?

Yes, for now. English v1.0 exams end on April 21, 2027 and other languages on October 21, 2027. CT-AI v1.0 certificates stay valid for life, like other Specialist certificates.

How long does CT-AI preparation take?

The syllabus sets 1,170 minutes (19.5 hours) of training for accredited courses, delivered over three days. Our five-week plan assumes about five hours a week of self-study.

Related certification guides

Return to the full certification directory →

Fees and booking

Use our exam fees overview to compare exam prices, taxes, retake charges and booking terms. Our country guides carry dated fee records where we have verified them. If your country is not listed, the provider guides cover international routes. Check the final amount and terms with your provider before you pay.