CT-AI v2.0 vs v1.0: What Changed in the 2026 Syllabus
ISTQB's Certified Tester AI Testing syllabus jumped to v2.0 in April 2026 — 11 chapters became 7, with new GenAI, LLM red teaming and dedicated data and model test levels. Here's every change and which version to sit.
ISTQB released the Certified Tester AI Testing (CT-AI) syllabus version 2.0 on 17 April 2026. The headline change: it shrank from 11 chapters to 7, was reorganized around the machine-learning lifecycle, added dedicated content on generative AI, LLMs and red teaming, and dropped the old chapter on using AI to improve testing. The exam itself is unchanged — 40 questions, 60 minutes, a 65% pass mark, with CTFL as a prerequisite.
What changed at a glance
If you studied for v1.0, most of the structure you remember has moved. The concrete differences:
Chapters: 11 reduced to 7.
Recommended training: 26 hours (4 days) down to 19.5 hours (3 days) — but the material is denser.
New test levels: input data testing and ML model testing are now treated as distinct levels.
New topics: GenAI and LLM testing, red teaming, retrieval-augmented generation (RAG), fine-tuning, and deployment testing (shadow and canary).
Removed: the chapter on using AI to improve testing, and AI test-environment considerations.
Standards: quality characteristics now map to ISO/IEC 25059, and the EU AI Act is referenced.
Exam format: unchanged — 40 questions, 60 minutes, 65% to pass, CTFL prerequisite.
From 11 chapters to 7
The v1.0 syllabus spread AI testing across 11 chapters that mixed concepts, techniques and tooling. v2.0 restructures the same ground so it follows the machine-learning lifecycle from data to deployed model:
Introduction to AI — AI vs conventional systems, the narrow/general/super AI spectrum, GenAI concepts, hardware and hosting, and the impact of regulation (EU AI Act).
Quality characteristics for AI-based systems — mapped to ISO/IEC 25059 (robustness, functional adaptability, controllability, intervenability).
The metrics and data that define ML performance — the foundation for judging a model.
Input data testing — a dedicated level focused on the data that trains and feeds a model.
ML model testing — a dedicated level for evaluating the trained model itself.
Testing risks across ML development and deployment — including shadow testing and canary releases.
A GenAI and LLM strand woven through the syllabus rather than bolted on at the end.
What's new in v2.0
The additions reflect what AI teams actually test in 2026:
GenAI and LLM testing: red teaming, prompt-based evaluation, and the challenge of non-deterministic text, image, audio and video outputs.
Input data testing as its own level: completeness, labeling quality, and data or target leakage.
ML model testing as its own level: functional-performance metrics such as
precision,recallandF1, plus metamorphic and back-to-back testing.Deployment testing: shadow testing, canary releases, and monitoring for data drift after go-live.
Modern practice: pretrained models, fine-tuning workflows and retrieval-augmented generation (RAG).
Standards alignment: the ISO/IEC 25059 AI quality model and references to the EU AI Act.
What was removed
The v1.0 chapter on using AI to improve testing activities — test case generation, defect prediction, regression optimization — has been dropped from the examinable content and pushed toward the generative-AI direction. Test-environment considerations specific to AI systems are also no longer examinable. If your old notes lean heavily on those topics, set them aside.
Which version should you take?
If you register now, you take v2.0 — it is the current General Availability syllabus. v1.0 is being retired: English-language exams remain available until 21 April 2027, and non-English exams until 21 October 2027. There is little reason to target the old version: v1.0 study material is largely outdated, and any employer or role that values the certification will expect the current syllabus. For the full picture of the exam, prerequisites and career value, see our ISTQB CT-AI certification guide.
How this changes your prep
Fewer hours does not mean less to learn — v2.0 is denser and more practice-oriented. Spend your time on the two new test levels (input data and ML model testing) and on the GenAI/LLM material, including red teaming, which barely existed in v1.0. Make sure you are comfortable reading a confusion matrix and computing precision, recall and F1, and understand where non-deterministic outputs break a traditional test oracle. CTFL remains a prerequisite, so the general testing vocabulary from the ISTQB Foundation Level still applies.
When you are ready to test yourself against v2.0-aligned questions, work through the ExamCaliber CT-AI mock exam — original questions with a rationale on every answer, covering GenAI, red teaming and the new data and model test levels. You can also consult the official ISTQB CT-AI certification page and the ISTQB glossary for exact terminology.
Frequently asked
The v2.0 syllabus reached General Availability on 17 April 2026.
Seven, reorganized around the machine-learning lifecycle — down from 11 in v1.0.
No. It is still 40 questions in 60 minutes, a 65% pass mark, with CTFL as a prerequisite.
English-language v1.0 exams are available until 21 April 2027, and non-English exams until 21 October 2027.
Dedicated GenAI and LLM testing (including red teaming), plus input data testing and ML model testing as separate test levels.
Yes. The ISTQB Foundation Level (CTFL) remains a prerequisite for CT-AI.
Part of the ExamCaliber editorial team. Every ExamCaliber question and rationale is written and reviewed by hand against the current syllabus — never scraped from exam dumps.