Log and document each incident with root cause: Which action best satisfies both aims?
Document each incident with cause and remediation and report serious incidents to the authority without undue delay.
The question
A provider of a systemic-risk general-purpose AI model learns that its model has repeatedly generated harmful outputs affecting users, and a regulator may inquire. Leadership wants a single approach that both meets its incident obligations and preserves evidence for later review. Which action best satisfies both aims?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Quietly patch the model and defer any external reporting until an internal review determines whether the outputs truly qualify as serious incidents.Plausible but wrong: for systemic-risk GPAI, serious incidents must be reported without undue delay, so deferring notification is not defensible.
- Publish a general transparency notice about model limitations, treating broad public disclosure as an adequate substitute for incident reporting.Plausible but wrong: a limitations notice is a transparency measure, not the structured incident documentation and authority reporting required.
- Route all affected outputs to human reviewers going forward, treating the added oversight as enough to close the incident duties.Plausible but wrong: adding oversight is a valid mitigation but does not fulfill the duty to document and report the incidents that already occurred.
- Log and document each incident with root cause and remediation, and report serious incidents to the relevant authority without undue delay. ✓Correct: documenting incidents with cause and remediation while reporting serious ones to the authority both meets obligations and preserves a defensible record.
The trap
Believing that mitigating or generically disclosing a problem discharges the duty to document and report incidents. How to remember it
Document each incident with cause and remediation and report serious incidents to the authority without undue delay.
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
- Red teaming that deliberately stress-tests the system: Which activity best fits this specific goal? →
- Model or data drift: Which contributing factor best describes this cause? →
- To describe the system's design: What is the primary purpose of this technical documentation? →
- 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.