Track performance against thresholds continuously: Which response best reflects that maintenance obligation?

Drift after release calls for continuous performance monitoring against thresholds plus scheduled retraining, which is the maintenance obligation.

The question

Preparing for AIGP? Take the free 5-min readiness quiz →

1. Recompute the pre-release conformity assessment once and re-issue the original model card without altering the model itself.

   Plausible but wrong: re-documenting release artifacts does not address the ongoing performance decay that retraining resolves.
2. Track performance against thresholds continuously and retrain on a scheduled cadence once drift breaches the defined limit. ✓

   Correct because III.C.2 requires continuous monitoring plus a regular schedule for maintenance, updates, and retraining to address drift.
3. Rely on user complaint tickets to signal decline and retrain only after a formal incident is opened by the support team.

   Almost right because complaints are one signal, but reactive-only retraining lacks the continuous monitoring and scheduled cadence required.
4. Freeze the model permanently to preserve its validated behavior and disclose the accuracy limitation to all downstream users.

   Plausible but wrong: freezing plus disclosure ignores the maintenance duty to update and retrain the system over time.

How to remember it

Drift after release calls for continuous performance monitoring against thresholds plus scheduled retraining, which is the maintenance obligation.

More Understanding How to Govern AI Development questions

• A model card summarizing intended use: Which readiness artifact directly addresses the standardized summary →
• Conduct scheduled red teaming and security testing: Which option best fits that periodic safety-assessment →
• Contain the harm: Which approach best satisfies incident management and documentation obligations under →
• All 426 Understanding How to Govern AI Development questions →