AI Use and Human Review Policy

01

Permitted AI assistance

• Organizing submitted records, extracting structured fields, translating text, summarizing timelines, and identifying missing materials.
• Preparing drafts for case managers and physician-facing case briefs that remain clearly marked as drafts until reviewed.
• Supporting workflow routing, document checks, and multilingual communication without making clinical decisions.

02

Prohibited uses

• AI must not independently diagnose, prescribe, select treatment, determine emergency severity, or guarantee an outcome.
• AI-generated text must not be represented as a licensed physician's final opinion.
• Patient information must not be used for model improvement unless separate, optional authorization and applicable safeguards are in place.

03

Required human review

• A case manager checks identity, source documents, missing information, translation context, and unsupported statements before patient-facing use.
• Formal clinical opinions and patient-specific recommendations require review by appropriately licensed physicians.
• Review actions, approvals, edits, and external sharing should remain traceable in the case audit trail.

04

Data minimization and analytics

• Public analytics must not receive names, contact details, diagnoses, medications, document names, case identifiers, or free-text medical content.
• Private patient, doctor, care-team, authentication, and demo workflow routes must not load public analytics tags.

05

Safety incidents and corrections

Suspected hallucinations, mistranslations, unsafe omissions, privacy exposure, or incorrect routing are escalated for human review. Affected drafts are withheld or corrected, the event is recorded, and recurring causes are addressed through workflow, prompt, model, or access-control changes.