Academic Medical Center: HIPAA-Compliant AI Deployment

The Challenge

By Q3 2024, the leadership at this leading Academic Medical Center faced a critical operational crisis. Physician burnout had reached an inflection point, driven largely by the "pajama time" phenomenon—clinicians spending hours after shifts completing EHR documentation. Internal surveys indicated physicians were spending 35% of their time on administrative tasks rather than patient care.

The Clinical Informatics team proposed deploying a cloud-based generative AI "ambient scribe" solution. However, the project was immediately halted by the Privacy Officer and CISO:

• The "Black Box" Problem: Cloud solutions could not cryptographically prove that patient data would not be used to train future models
• HIPAA & Cloud Egress: Transmitting PHI to third-party API endpoints created an expanded attack surface
• ONC HTI-1 Regulations: New transparency requirements for "Decision Support Interventions" that commercial LLMs could not satisfy

The PRYZM Solution

PRYZM deployed a Sovereign AI instance directly within the AMC's existing secure infrastructure using AWS Nitro Enclaves:

• Cryptographic Attestation: Hardware-level attestation proves that the code running in the enclave is exactly what was audited
• Data Isolation: PHI is processed entirely in volatile memory within the enclave. Once the clinical note is generated and pushed to the EHR, the data is cryptographically shredded
• Evidence Packs: Cryptographically signed, immutable audit trails generated for every patient encounter—capturing raw input, model version, prompt logic, output, and physician edits

The system acts as a drafter, not an autonomous diagnostician. Every recommendation is hyperlinked to the specific line in the patient transcript that generated it, keeping the human firmly in the loop and satisfying FDA CDS guidance Criterion 4.