A policy and control to deactivate the AI system so it can: Which pre-established control most directly
An immediate unsafe-behavior incident calls for a pre-built control to deactivate and take the system offline.
The question
A deployed medical-chatbot begins giving unsafe dosage suggestions during a live update, and the team must stop it acting on users immediately while they investigate. Which pre-established control most directly enables this?
Preparing for AIGP? Take the free 5-min readiness quiz →
- A policy and control to deactivate the AI system so it can be taken offline immediately when unsafe ✓A deactivation control lets the team pull the system offline at once, halting harm while the root cause is investigated.
- A quarterly audit schedule that periodically reviews the chatbot's aggregate accuracy and safety metricsA periodic audit surfaces trends later but cannot stop an acute unsafe behavior happening right now.
- A vendor indemnity clause covering financial losses that arise from the model's erroneous outputsIndemnity addresses after-the-fact liability, not the immediate need to stop unsafe outputs.
- A user-training module explaining how staff should phrase safe questions to the chatbot properlyUser training shapes inputs but cannot halt the system's dangerous outputs during an active incident.
The trap
Choosing a detective or contractual control when an acute safety incident requires an immediate deactivation capability. How to remember it
An immediate unsafe-behavior incident calls for a pre-built control to deactivate and take the system offline.
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 Deployment and Use questions
- An external communication plan assigning ownership: Which governance element does this describe? →
- Ethical considerations: Which use-case context factor most directly frames this concern in the deployment →
- A classic classification model trained to assign claims: Considering model-type differences, which choice best →
- All 424 Understanding How to Govern AI Deployment and Use questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Deployment and Use ·
Every answer, right and wrong, comes with its own explanation.