Conduct scheduled red teaming and security testing: Which option best fits that periodic safety-assessment
Proactively probing a deployed model for unsafe or exploitable behavior is red teaming and security testing, a periodic safety-assessment activity.
The question
A governance committee wants periodic, adversarial evaluation of a deployed generative model to surface prompt-based jailbreaks, unsafe outputs, and security weaknesses before they cause harm. They ask which activity best provides this proactive stress test. Which option best fits that periodic safety-assessment need?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Increase the sampling rate of routine accuracy dashboards so that average output quality is tracked more frequently over time.Plausible but wrong: higher-frequency accuracy monitoring measures typical performance, not adversarial failure modes that red teaming targets.
- Refresh the training-data summary and copyright policy so the disclosed provenance keeps pace with new sources added.Almost right because documentation matters, but data-summary updates are transparency artifacts, not adversarial safety testing.
- Conduct scheduled red teaming and security testing that adversarially probes the model for unsafe or exploitable behavior. ✓Correct because III.C.3 names red teaming, threat modeling, and security testing as periodic activities to assess performance, reliability, and safety.
- Expand the user instructions-for-use document to warn operators about the categories of prompts that may yield unsafe replies.Plausible but wrong: user guidance is a transparency measure and does not itself test the model for exploitable weaknesses.
The trap
Assuming intensified routine monitoring or documentation substitutes for adversarial red teaming. How to remember it
Proactively probing a deployed model for unsafe or exploitable behavior is red teaming and security testing, a periodic safety-assessment activity.
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
- Track performance against thresholds continuously: Which response best reflects that maintenance obligation? →
- Contain the harm: Which approach best satisfies incident management and documentation obligations under →
- Data drift, where production input distributions diverge: Which factor does the evidence most directly →
- 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.
