The RAI Test

Question 1

Under the EU AI Act, an AI system that is a medical device (regulated under the EU Medical Device Regulation) is classified as high risk through Annex I. After the 2026 Digital Omnibus, when do the high-risk obligations for these product-embedded systems apply?

Accepted Answer

August 2, 2028

Question 2

Six months after a hospital deploys a deterioration prediction model, the distribution of incoming lab values shifts because the lab changed analyzers. Outcome labels will not be available for another 90 days. What should the monitoring team do now?

Accepted Answer

Track input data drift (for example, population stability index or a Kolmogorov-Smirnov test) as a leading indicator, check whether alert rates changed across patient subgroups, and escalate under predefined thresholds

Question 3

A 78-year-old man with end-stage kidney disease declines to start dialysis. He says he understands he will likely die within weeks, that he has seen what dialysis did to his brother, and that he would rather spend his remaining time at home. His daughter insists the team start dialysis anyway. What is the most ethically appropriate response?

Accepted Answer

Confirm his decision-making capacity (he understands the facts, appreciates how they apply to him, reasons in line with his values, and expresses a consistent choice), then respect his refusal while offering palliative care and support for the family

Question 4

A hospital AI allocates scarce ICU beds by strictly following a published triage policy designed to save the most lives across all patients. In one case, breaking the policy would save an extra life, but the system follows the policy anyway because consistent rules produce the best outcomes over time. Which ethical framework does this design reflect?

Accepted Answer

Rule utilitarianism

Question 5

A runaway trolley will kill five workers on the main track. You can pull a lever to divert it onto a side track, where it will kill one worker. What do you do?