ISTQB Specialist (CT-GenAI v1.1) Probeprüfung #4

The fourth full-length practice exam for the ISTQB Certified Tester Testing with Generative AI (CT-GenAI) v1.1 syllabus. It follows the official exam format: 40 questions worth 46 points, 60 minutes, and 30 points to pass (65%). Scoring is weighted the way the real exam weights it: K1 and K2 questions are worth 1 point and K3 questions 2, so the six applied questions carry 12 of the 46 points between them. The chapter split is the official one: 7 questions from Introduction to GenAI for Software Testing, 11 from Prompt Engineering for Effective Software Testing (16 points, and the home of five of the six K3 items), 10 from Managing Risks of GenAI in Software Testing, 5 from LLM-Powered Solutions for Software Testing and 7 from Deploying and Integrating GenAI in Test Organizations. The K-level mix matches the published structure as well: 8 questions at K1, 26 at K2 and 6 at K3. Mock exams 1 to 3 concentrate on prompting for test analysis and test data, on model behaviour and operational risk, and on making generated work checkable. This one looks at model choice, automation work and agents. The first chapter asks how kinds of model differ in practice: base and instruction-tuned models, reasoning models, sensitivity to wording, the tendency to agree with the user, and API use against chat use. The prompting questions move into automation work: separating instructions from pasted input, coding conventions for generated scripts, reusing an existing BDD step library, impact analysis after a rule change, a test summary report grounded in real figures, invalid input derived from a validation rule, converting manual cases to Gherkin, and an accessibility review of form markup. The risk chapter covers data poisoning, data minimisation, the energy footprint of GenAI, reasoning errors in worked answers, instructions hidden in retrieved content, how the EU AI Act assigns risk classes, which governance framework can be certified, and bias carried in by few-shot examples. The solutions chapter deals with an agent that meets its goal the wrong way, how tool calling works, least privilege for a connected assistant, pinning model versions in CI and trusting model-graded evaluation. The last chapter covers AI literacy, total cost of ownership, tool selection criteria, AI champions, generated code in the pipeline, pilot results that argue against scaling, and resistance from experienced testers. Every option, correct and incorrect alike, carries its own rationale, and every question has an overall explanation. As with the other mocks, remember that CT-GenAI is about testing with generative AI; testing AI-based systems is the subject of CT-AI, which is a separate exam.