Security Overview

Introduction¶

The JupyterHealth Pre-MVP security focuses on establishing a secure privacy-preserving transmission of patient data from:

API Credentials¶

• CommonHealth Application will authenticate with the Medical Device API using:
  • Client ID
  • Client Secret
• Each Partner will authenticate with the CloudStorage Service using:
  • Partner ID
  • Client ID
  • Client Secret

API credentials in JupyterHub¶

The JupyterHealth pre-MVP application is registered as a Partner application (two, actually - one ‘prod’ to represent the pre-MVP itself, and one ‘testing’ for testing configurations).

There are two partner applications registered:

• ‘prod’ represents the pre-MVP
• ‘testing’ for testing out functionality

These credentials are stored in AWS SecretManager:

• ch-cloud-creds-{name} contains the partner id, client id/secret, etc. required to access the CommonHealth Cloud API
• ch-cloud-storage-{name} contains the state of the CH Cloud “storage delegate”. This is the local storage of:
  • decryption keys for patient data
  • patient id/uuid mapping

The JupyterHub user ‘role’ is granted read access to ch-cloud-creds-{name} and read/write access to ch-cloud-storage-{name}.

Pre-MVP considerations¶

In the pre-MVP, from the perspective of CommonHealth cloud and data access, all JupyterHub users are equivalent and indistinguishable and act collectively as the pre-MVP. That is, all JupyterHub users have:

• read access to the client id/secret to access the API in order to:
  • create new deep links to register users
  • fetch encrypted patient data
• read/write access to the decryption key for decrypting patient data
• read/write access to the patient id mapping to identify patients and store new patient id mappings

Following the pre-MVP, users will login to JupyterHub with SMART-on-FHIR as an OAuth provider, and will authenticate directly as themselves via scoped access tokens, rather than always acting as the partner application. This will allow per-user access control, registered and enforced by SMART-on-FHIR, outside JupyterHub.

Encryption¶

• Each Partner will generate a set of Private/Public keys that will used for encryption and decryption of patient data.
• The Partner Public encryption key will be stored in the CloudStorage Service.
• The CommonHealth Application will use the Partner’s Public Key to encrypt the patient data before storing it in the CloudStorage Service.
• The Partner’s Private Key will be used to decrypt the patient data when retrieved from the CloudStorage Service.

Cryptography¶

• Google Tink is utilized for client-side as well as server-side cryptography.

• The following primitives are required:

  • DataGroup encryption key pair: HybridEncryption using the following algorithms: ECIES_P256_HKDF_HMAC_SHA256_AES128_GCM

  • EncryptedRecord DEK: AEAD using the following algorithms: AES256_GCM

  • Signing key(s) for TrustedPartners: ECDSA_P256

Data Flow Diagram¶