Design for ePHI handling
Provider eligibility and telemetry decisions are explicit, but retention enforcement and regional/dedicated cells remain required work.
For healthcare teams
Control AI paths before sensitive workflows depend on them.
The architecture limits stored metadata and is designed with later BAA and dedicated-deployment needs in mind. No compliance status is claimed before the required agreements and reviews are complete.
No prompt persistence by default
Tenant-bound provider credential encryption
Access, audit, incident, and risk-analysis foundations
In plain language
Design the AI path so later assurance does not require rebuilding the product.
Healthcare readiness is more than encryption or a badge. The service begins with minimized hosted telemetry, tenant-bound credentials, explicit workload/data context, and provider eligibility controls. A BAA, risk analysis, retention validation, workforce procedure, vendor review, and deployment boundary are still required before ePHI use.
The readiness outcomeA non-sensitive pilot can validate the control model now while keeping regulated production outside the boundary until administrative and technical prerequisites are complete.
For the operating team
Make every regulated dependency explicit.
Treat provider contracts, data class, region, retention, support, incident response, and deployment model as release gates.
What administrators configure
Workload classification
Require an explicit data class and prohibit regulated traffic from unapproved routes.
Provider boundary
Limit credentials and models to reviewed accounts, contracts, regions, and telemetry behavior.
Administrative gate
Document BAA, risk analysis, procedures, training, incident handling, and validation before ePHI.
What happens to each request
- 01Classify
Use explicit workload and data-class context; never infer authorization from a model name alone.
- 02Restrict
Limit eligible providers, models, credentials, and telemetry before the request can leave the gateway.
- 03Operate
Run the tested rule in Observe, Shadow, or Enforce according to the approved rollout stage.
- 04Review
Retain metadata evidence and administrative change history for authorized operational review.
See the working control
Calibrate local protection before enforcing it for healthcare safeguards.
For healthcare safeguards, versioned profiles screen prompt attacks before spend and sensitive output before release. They also limit the tool names, types, and MCP hosts a model may use. Schemas, arguments, and results are not retained.
- This healthcare safeguards view runs in the working customer console
- Exercised with safe OpenAI- and Anthropic-compatible simulators
- The healthcare safeguards capture excludes keys, prompt text, responses, and private email
For healthcare safeguards, versioned profiles screen prompt attacks before spend and sensitive output before release. They also limit the tool names, types, and MCP hosts a model may use. Schemas, arguments, and results are not retained.
What administrators can change
The settings behind for healthcare teams.
For healthcare teams, the console shows each setting, its owner, and its scope so administrators can review changes before they reach live traffic.
Keep keys customer-controlled
Customers bring provider credentials today. A release-candidate buyer-owned AWS Bedrock data plane is limited to a documented us-east-1 boundary; customer-managed KMS and regulated-workload approval are not included.
Administrative readiness
BAA execution, risk analysis, policies, workforce procedures, vendor review, and validated operations are required before any HIPAA use.
Evidence, not assertions
Build evidence now without claiming compliance early.
The trust center separates implemented controls, deployment dependencies, unavailable capabilities, and assurance that still requires an independent review.
Review security boundaries and current status- Workload classification: Require an explicit data class and prohibit regulated traffic from unapproved routes.
- Provider boundary: Limit credentials and models to reviewed accounts, contracts, regions, and telemetry behavior.
- Administrative gate: Document BAA, risk analysis, procedures, training, incident handling, and validation before ePHI.
Questions teams ask
Questions about for healthcare teams.
What buyers, administrators, and security reviewers need to confirm for for healthcare teams.
Can we send PHI through the hosted service today?
No. AI Gateway HQ does not currently claim HIPAA readiness or offer a BAA. Use only non-sensitive test data until contractual, administrative, and technical gates are completed.
Are prompt bodies retained?
The hosted gateway is designed not to persist prompt or response bodies by default. Provider retention and application logging remain separate customer review responsibilities.
Will customer-VPC deployment be supported?
A release-candidate buyer-owned AWS package now implements a private-VPC Bedrock path, paid-period entitlement, read-only verification, buyer-approved update, and removal flow. It is not approved for ePHI and remains gated on a fresh external acceptance run before general availability.
Start safely
Start guided setupSee what the policy would do before it can block production.
Start with a provider connection, a limited workload key, and Observe mode; move to enforcement after your team confirms the result.