Adversarial Example
An adversarial example is specially crafted input designed to cause an artificial intelligence or machine learning model to produce an incorrect prediction or decision.
“An input to an ML model created by applying small perturbations to a working example that results in the model outputting an incorrect result with high confidence.”
(Definition reproduced from the ISTQB Glossary. Copyright belongs to ISTQB.)
View the complete ISTQB GlossaryWhat is Adversarial Example?
Adversarial examples exploit weaknesses in AI models by making small, often imperceptible, changes to input data that significantly affect model output.
Purpose: Reveal weaknesses in machine learning models.
Creation: Inputs are intentionally modified to influence AI predictions.
Testing value: They help evaluate the robustness of AI systems against manipulation.
Real World Example
An online banking login flow is being reviewed after fraud analysts report repeated attempts to discover valid customer accounts.
The concern is not whether ordinary login works; it is how the system behaves when someone intentionally misuses it. Adversarial example points testing toward that hostile or risky behavior.
The tester probes rate limits, error messages, audit events, lockout behavior, and monitoring signals to see whether the system gives attackers useful feedback.
The team strengthens the controls because Adversarial example helps them test the product from the perspective of misuse, not only normal customer journeys.
Practice Questions
Question 1
What is an adversarial example in the context of ML model testing?
Question 2
Why are adversarial examples particularly concerning for ML models in safety-critical systems?
Test your knowledge with real ISTQB-style questions
You’ve learned Adversarial Example. Test your understanding with topic-specific questions in our Mock Exams.