ExamworthyExamworthy
Open the appStart free

AIGP - Understanding how to govern AI deployment and use - Section 4.5

Apply policies, best practices and ethical considerations to the deployment of an AI system, covering data governance, risk management, user training, continuous monitoring, and periodic audits.

Apply AI deployment governance policies covering data governance, risk management, user training, continuous monitoring, and periodic AI audits. Recognise how each control addresses specific operational risks that emerge once an AI system is in production.

AI deployment governanceuser trainingcontinuous monitoringperiodic AI audit

Practice question for this objective

Free sampleUnderstanding how to govern AI deployment and usehard

A logistics firm has a deployed AI system that predicts next-day demand from a live feed of order, weather and traffic data supplied through a third-party integration. The system passed data-quality validation at launch. Six months later its forecasts have degraded, and investigation shows the upstream provider quietly changed how it codes one weather field, so the live inputs no longer match the format the system was trained on. The operations team wants a data-governance control that would catch this class of problem during operation. Which control most directly addresses it?

  • ARe-run the original launch-time data-quality validation suite once a year and compare the results against the launch report
  • BEncrypt the third-party data feed in transit and require the provider to authenticate before each upload to the prediction system
  • CMaintain a documented lineage record of which datasets, sources and consent conditions were used to train the original demand system
  • DContinuously validate incoming production data against the expected schema and statistical profile of the training distribution, and alert when live inputs deviate Correct
Recognise that data governance in operation requires continuous validation of production inputs against the training schema and distribution, since upstream feeds can silently change after launch. Data governance across the AI life cycle does not end at launch validation; a deployed system depends on live inputs that third-party providers can change without notice. Continuously checking incoming production data against the expected schema and statistical profile detects such a shift at the input stage, before it propagates into degraded outputs. Annual re-validation is too infrequent, encryption secures the channel rather than the data's fitness, and lineage documents past training inputs rather than monitoring the live feed.

Why A is wrong: An annual re-validation is better than nothing, but a yearly cadence would let a silent upstream format change degrade forecasts for months before detection, so it does not address an input feed that can shift at any time.

Why B is wrong: Transport encryption and authentication are sound security controls and may be required anyway, but they protect the feed's confidentiality and integrity in transit, not whether the data's coding and distribution still match what the system expects.

Why C is wrong: Training-data lineage is essential for tracing what a system learned from and for impact analysis, but it is a static record of past inputs and does not monitor whether today's live feed has drifted from the expected format.

Why D is correct: Ongoing input-data validation against the training schema and distribution is the data-governance control that catches upstream changes as they occur, because it tests the live feed itself rather than only the system's downstream predictions.

See more AIGP practice questions, answers explained.

More in this domain

Back to all Understanding how to govern AI deployment and use objectives, or the AIGP cert hub.

Examworthy is not affiliated with or endorsed by IAPP. Original, blueprint-aligned practice material only.