Skip to content
ushealthcloud

Trust Center · HIPAA

HIPAA-compliant storage and processing

How we store and process protected health information (PHI) in line with the HIPAA Privacy and Security Rules, including when that data is used to train AI models.

Scope

This page covers how UsHealthCloud stores and processes PHI on behalf of the healthcare organizations we work with, across data science, AI automation, cybersecurity and AI model engagements.

Controls in every engagement

Four controls are mandatory in every engagement: encryption in transit and at rest, least-privilege role-based access, tamper-evident audit trails, and human approval for any AI action that affects a clinical or coverage decision.

Storage and hosting

PHI is stored and processed only on HIPAA-eligible services of a major cloud provider, under a signed Business Associate Agreement with that provider.

US customer data stays in US regions. EU data stays in an EU region unless a documented transfer mechanism applies.

Encryption and keys

Data at rest is encrypted with AES-256, covering storage, databases, backups and model artifacts, in line with NIST SP 800-111. Data in transit, including traffic between internal services, uses TLS 1.2 at minimum and TLS 1.3 where supported, in line with NIST SP 800-52 Rev. 2. Encryption runs in FIPS 140-validated cryptographic modules.

Keys are held in a cloud key management service backed by FIPS 140-validated hardware security modules, with separate keys for each customer environment and automatic annual rotation. Customer-managed keys are available on request. Only named administrators can use keys, and every use is logged.

Access control and audit

Access to PHI follows the minimum necessary standard: identifiable data is restricted to named roles for specific tasks, and de-identified data is used wherever the work allows.

All access to PHI, encryption keys and administrative consoles is logged to tamper-evident storage and kept for six years. Automated alerts are triaged daily, our security lead reviews access logs monthly, and user access rights are recertified every quarter.

PHI and AI model training

Models are trained or fine-tuned on de-identified data by default. Identifiable data is used for training only when the Business Associate Agreement or contract expressly permits it and the use is allowed under HIPAA.

De-identification follows one of the two HIPAA methods in 45 CFR 164.514(b): Safe Harbor (removal of 18 specified identifiers) or Expert Determination.

Each customer's data is logically isolated and used only for that customer's contracted work. It is never pooled or used to train models for other customers.

Business Associate Agreements

We sign a Business Associate Agreement (BAA) with every customer whose PHI we handle.

Each BAA covers permitted uses and disclosures, the safeguards we maintain, breach notification (with timing set in the BAA and never beyond HIPAA's 60-day limit), BAAs with any subcontractors who handle PHI, and the return or destruction of PHI when the engagement ends.

Related

FAQ

HIPAA questions

  • CertificationIs UsHealthCloud HIPAA certified?

    There is no official HIPAA certification. HHS does not certify organizations or products. We describe our practice precisely: HIPAA-compliant storage and processing.

  • Training dataWill our PHI be used to train models for other customers?

    No. Each customer's data is logically isolated and used only for the work set out in your agreement with us. It is never pooled with other customers' data or used to train models for them.

  • AgreementsWill you sign a Business Associate Agreement?

    Yes. We sign a BAA with every customer whose PHI we handle, before any PHI is shared.

Next step

Need more detail for a security review?

Tell us what your assessment asks for, and we'll respond with specifics.

What the first call gives you

  1. 1A map of where PHI flows today
  2. 2One workflow worth starting with
  3. 3The controls that workflow needs

A short scoping call, then a written proposal. Please don't send patient information.