You are deploying an AI-driven applicant-screening agent that analyzes candidate resumes and social-media data to...

The selected option specifically B states “Pseudonymize protected attributes, implement fairness-aware debiasing, maintain an audit trail, and enforce GDPR data-minimization and consent.”, which matches the operational requirement rather than a superficial wording match. Pseudonymization, fairness-aware debiasing, audit trails, consent, and data minimization address both discrimination and GDPR obligations. Encryption alone is incomplete. The architecture implied by Option B is the one that survives real workloads: separate responsibilities, explicit contracts, and measurable runtime behavior. In NVIDIA terms, NeMo Guardrails adds programmable controls around LLM applications, can wrap LangChain flows, and supports policy checks before and after model/tool execution. The practical pattern is responsible AI controls that are part of the runtime path, not just model-card language or prompt reminders. That is why the other options are traps: authentication tells you who used the system; it does not prove the generated content stayed compliant. This is exactly where NVIDIA’s stack is strongest: separating acceleration, orchestration, policy, and observability.