Question

A healthcare organization is deploying an AI-powered patient triage assistant. During a responsible AI review, the team discovers that the solution logs full patient conversation transcripts—including names, diagnoses, and contact details—in plaintext to an unencrypted analytics store accessible by the entire product team. Which Microsoft responsible AI principle is most directly violated, and what should the AI council require before approval?

Accepted Answer

Privacy and security — require data minimization, encryption, and role-based access controls before deployment.

Answer

Fairness — require the team to test the model across demographic groups to ensure equitable triage outcomes.

Answer

Transparency — require the team to publish a model card explaining how the triage recommendations are generated.

Answer

Reliability and safety — require additional load testing to ensure the assistant responds within acceptable time limits.