Require scoped evaluation and approval before repurposing: What is the defensible next step?
A new access-affecting purpose lacks predictive evidence, so scoped evaluation and approval must precede repurposing chatbot conversations.
The question
A travel-support chatbot was approved to answer itinerary questions using retrieved policy pages. Operations now proposes using its conversations to prioritize passengers for proactive assistance. No evaluation shows that conversation signals predict need, and the new ranking could affect access to limited support. What is the defensible next step?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Deploy ranking temporarily while collecting passenger feedback.Feedback during deployment does not substitute for prior evidence when an unvalidated ranking may influence limited assistance allocation.
- Require scoped evaluation and approval before repurposing conversations. ✓The secondary use changes purpose and may affect access; scoped evaluation, review, and approval are needed before operational deployment.
- Pilot the ranking using existing chatbot confidence scores.Confidence scores do not establish need prediction or fairness for access prioritization, especially when the proposed use differs from the approved purpose.
- Reuse the original approval because the same chatbot generates both outputs.Shared technology does not make purposes equivalent; the new decision context creates distinct performance, impact, and oversight requirements.
The trap
A new purpose can create new risks even when the same model and data are involved; assess purpose, impact, evidence, and authority separately. How to remember it
A new access-affecting purpose lacks predictive evidence, so scoped evaluation and approval must precede repurposing chatbot conversations.
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
- Measure performance for the affected local population: What should the monitoring team do first? →
- Revoke the messaging permission and preserve action logs: Which response best addresses the changed operating →
- Treat retrieved instructions as untrusted data: What operating assumption must change? →
- 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.
