Sensitive casework must use private or hosted models, not public consumer LLM endpoints. Which policy distinction is this?

Policies often require private or organizationally hosted models for sensitive workloads while restricting public consumer LLM endpoints. Applying that private-versus-public distinction to municipal casework keeps confidential case notes off free public chat services whose retention, training use, and subprocessors are not under city control. A prompt-logging control policy keeping sensitive municipal casework off consumer LLMs gets the control type wrong; prompt logging addresses whether prompts are stored or audited, not the private-versus-public distinction about where sensitive casework may be processed.

A data-retention limit policy keeping sensitive municipal casework off consumer LLMs gets the policy type wrong; it addresses how long data is retained, not the private-versus-public distinction about where sensitive casework may be processed. A prompt-logging control policy keeping sensitive municipal casework off consumer LLMs fails the distinction test: it focuses on logging of prompts while leaving unanswered the private-versus-public question of where sensitive casework may be processed. An output-review control policy keeping sensitive municipal casework off consumer LLMs gets the control type wrong; it addresses human review of outputs, not the private-versus-public distinction about where sensitive casework may be processed.

Exam caveat: private-versus-public model policy is about where sensitive casework may be processed—not about prompt-logging controls for sensitive casework. Adding output review does not satisfy the required private-versus-public distinction about where processing may occur.

Operational check: route sensitive casework only to private/hosted sanctioned models, disable or warn on public consumer endpoints from caseworker workflows, and verify DLP rules catch case-note exfiltration to public LLMs.