Security testing that probes the pipeline and model: Within the training-and-testing plan, which test type
Security testing targets adversarial manipulation such as data poisoning, directly addressing maliciously injected samples.
The question
A team ingests training data from an open community feed and worries that an attacker could inject manipulated samples to corrupt the model's behavior. Within the training-and-testing plan, which test type most directly targets this threat?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Validation testing on a clean holdout set that confirms the model's accuracy generalizes beyond the training data it was directly optimized upon.Validation checks generalization on clean data; it can miss poisoned behavior that only triggers on attacker-chosen inputs.
- Security testing that probes the pipeline and model for data-poisoning and other adversarial manipulation, verifying resistance before the model is trusted. ✓Security testing is the plan element aimed at adversarial threats such as data poisoning, directly addressing the manipulated-sample concern.
- Interpretability testing that makes individual predictions explainable to reviewers, so the reasoning behind each automated decision can be understood and challenged.Interpretability aids understanding of decisions but is not designed to detect deliberately injected poisoned training samples.
- Performance testing that measures throughput and latency under production load, confirming the system responds quickly enough in operation.Performance testing measures speed and scalability, which has no bearing on resistance to data-poisoning attacks.
The trap
Assuming clean-data validation or performance testing will catch maliciously injected training samples. How to remember it
Security testing targets adversarial manipulation such as data poisoning, directly addressing maliciously injected samples.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Development questions
- The origin of each portion: To satisfy data lineage and provenance requirements, what must the provenance →
- Investigate the disparity's source in data or design: Under III.B.4, what is the most appropriate way to →
- A test record listing each test type performed: Applying III.B.5, which documentation practice would let the →
- All 426 Understanding How to Govern AI Development questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Development ·
Every answer, right and wrong, comes with its own explanation.
