ISTQB CT-AI (Certified Tester AI Testing): The Complete 2026 Guide

The ISTQB CT-AI (Certified Tester AI Testing) certification validates that a tester can systematically test AI-based systems — especially machine learning (ML) and generative AI such as large language models. The current syllabus, version 2.0, became available on 17 April 2026 and is the version every new candidate should study. In short: 40 questions worth 44 points in total, 29 points to pass, 60 minutes, CTFL required — the full breakdown, including what ISTQB does and does not publish about this exam, is in our CT-AI v2.0 exam guide.

What is the ISTQB CT-AI certification?

CT-AI is a vendor-neutral specialist certificate in the ISTQB scheme. It does not teach you to build models; it teaches you to test them — to find where an AI system behaves incorrectly, unfairly, or unpredictably, and to judge its quality with the right metrics. It sits above the Foundation Level (CTFL) and targets testers, test analysts, test engineers, and the data scientists and analysts who want to understand how their data and models will be validated.

What is new in CT-AI v2.0 (2026)?

Version 2.0 is a substantial rewrite, not a refresh. The headline changes are:

• Reorganised into seven lifecycle-based areas, ordered by input data, the model, and the ML development pipeline rather than by AI theory.

• Generative AI and LLM testing added: prompt robustness, hallucination and groundedness, non-determinism, and red teaming.

• Two explicit ML test levels: input-data testing (representativeness, label quality, leakage) and model testing (metrics, metamorphic and back-to-back testing).

• Deployment testing, including data drift and rollout strategies such as canary releases.

• Modern model practice: pretrained models, fine-tuning, and retrieval-augmented generation (RAG).

• The "using AI for testing" block was removed and now lives in the separate CT-GenAI certification.

The material is denser and more hands-on than v1.0, so treat a v1.0 course or question bank as out of date rather than as a shortcut. Our detailed CT-AI v2.0 vs v1.0 comparison walks through every change.

CT-AI or CT-GenAI? They are not the same exam

This is the most common booking mistake in the ISTQB catalogue, and it costs a full exam fee. CT-AI is about testing AI: the AI system is your test object, and your job is to sign off on a model that somebody else trained. CT-GenAI (Certified Tester Testing with Generative AI) runs in the opposite direction — using generative AI as a tool for your own testing work, from test ideas to test data. If you have to approve a classifier, a recommendation engine or an LLM feature, you want CT-AI. If you want an assistant that writes test cases faster, you want CT-GenAI. And if you want both, the order is clear: CT-AI teaches you how a model fails, and that scepticism is exactly what you need to use generative AI responsibly.

What the CT-AI v2.0 syllabus covers

Quality characteristics of AI systems

AI systems introduce quality factors that classic software does not have: bias and fairness, transparency and explainability, autonomy, adaptability, and non-determinism. A correct-looking output can still be wrong if the model is biased or the result cannot be explained. CT-AI teaches you to treat these as first-class test objectives.

Testing machine learning systems

This is the core of the exam. You test the data before the model (completeness, representativeness, label correctness, leakage) and then the model itself using metrics such as precision, recall, and F1 read from a confusion matrix. Techniques include metamorphic testing, back-to-back testing against a reference implementation, and A/B testing in production. Understanding why accuracy alone is misleading on imbalanced data is a frequent exam theme. If those metrics are new to you, start with our walkthrough of precision, recall and F1 for testers.

Testing generative AI and LLMs

The newest material. Because LLM output is non-deterministic, you cannot rely on a single expected result; instead you test for groundedness, factuality, safety, and robustness against adversarial prompts. Red teaming — deliberately probing a model for unsafe or policy-breaking behaviour — is now part of the syllabus, alongside RAG-specific risks and evaluation metrics.

Prerequisites and who should take it

The only hard prerequisite is CTFL. Beyond that, CT-AI suits testers moving into AI projects, automation engineers validating ML pipelines, and data scientists who want a shared vocabulary with QA. You do not need to be able to train a model from scratch, but comfort with basic statistics and the idea of a confusion matrix will help.

And who it is not for: if nobody in your organisation ships a model and no such project is in sight, the certificate will not pay for itself this year — CTFL plus a specialism in test design techniques returns more for the same effort. CT-AI pays off once you are the person who has to say whether a probabilistic system is good enough to release, because that is a question classic test design does not answer.

What the certificate is worth in practice

Two honest expectations. On its own it will not get you an AI-testing job: interviews in this field are about confusion matrices, data leakage and evaluation design, and the certificate only shows that you know the vocabulary. What it does deliver is a defensible, vendor-neutral framework in a field where nearly every alternative is a tool vendor's course, and a shared language between QA and data science — which in practice is where AI projects lose the most time. If your employer pays for accredited training, the structure it imposes on a messy subject is worth more than the badge.

How to prepare

Start from the official CT-AI syllabus and the ISTQB glossary, then drill the harder K3 areas — metric calculation, input-data defects, and LLM evaluation — with realistic practice questions rather than rote definitions. Our free CT-AI mock exam mirrors the real format (40 questions, 65% pass) and includes scenario and confusion-matrix items with a written rationale on every answer, so you learn why an option is right or wrong. For a week-by-week plan there is our CT-AI study plan, and the reference facts — points, timing, the non-native-language allowance — live in the CT-AI v2.0 FAQ.

If you have not yet earned CTFL, start with our complete ISTQB CTFL Foundation guide — it is the gateway to CT-AI and to the wider ISTQB path.