HIPAA on AWS

On AWS, the account boundary does most of the HIPAA work

Amazon Web Services (AWS) lets any service run in an account covered by its Business Associate Addendum, but asks that PHI stay in the services it lists as HIPAA eligible. That gap is where most risk lives. The cleanest way to close it is structural: put ePHI in accounts built for it.

Back to HIPAA services

01 · Worked scenario

Illustrative: a telehealth start-up with a mixed stack

Trace the patient data first. Then decide where each piece may live.

Illustrative only. A start-up sells video visits to US clinics. Its booking app, database and file store run on AWS. Video sessions use a third-party video provider, error tracking is a separate tool, and the product team wants dashboards on visit data. Everything began life in one AWS account shared with marketing.

ComponentHolds ePHI?What decides itAction
Booking databaseYes: names, visit reasonsAWS addendum, eligible serviceMove to the ePHI account; encrypt with a managed key
Uploaded documentsYes: referrals, photosAWS addendum, eligible servicePrivate storage, versioned, access logged
Video sessionsYes, in transitThe video vendor's own BAASign it, or change vendor
Error trackingOften, by accidentThat tool's BAA, if anyScrub payloads, or sign a BAA
Product dashboardsOnly if fed raw recordsYour data designFeed de-identified or aggregate data
Marketing siteNoNothingKeep it out of the ePHI account

Only two rows are settled by AWS at all. The others depend on other vendors and on design choices.

02 · The addendum

Accepting the Business Associate Addendum

Accepted in AWS Artifact, for one account or a whole organisation.

AWS calls its agreement the Business Associate Addendum and handles it in AWS Artifact, the console area for agreements and audit reports. You can accept it for a single account, or, from the management account of AWS Organizations, for every account in the organisation, including accounts created later.

Account-level acceptance

  • Covers the one account that accepted
  • Good when only one account will hold ePHI
  • Easy to miss when a new account is created

Organisation-level acceptance

  • Covers existing and future member accounts
  • No gap when teams add accounts
  • Every account is then treated as able to hold PHI, so guardrails matter more

Read the addendum's current terms in Artifact before you accept. AWS can change them, and accepting is a legal act for your company.

03 · Eligible services

Eligible services, and everything else in the account

The addendum allows any service in the account. PHI may only go into the eligible ones.

AWS publishes a reference list of HIPAA eligible services and adds to it over time. The list changes, so we do not reproduce it here. Read the current one before each design decision, and record which version you checked.

  • A service that is not on the list can still run in the account. It must never receive PHI: no patient fields in its inputs, names, tags or logs.
  • Metadata leaks too. Object keys, queue names and resource tags built from patient names put PHI where you did not plan it.
  • A new managed service that the team wants to try needs a check against the list first, not after the pilot.

04 · Account structure

Give ePHI its own account

An AWS account is a hard boundary. Use it as one.

Mixing patient data with marketing, experiments and build tooling in one account means every person and role in that account is in scope. Separate accounts shrink the scope and let you apply stricter rules where ePHI lives without slowing everyone else down.

  1. ACCT.1
    ePHI workload accountProduction systems that store or process patient data, and nothing else.
  2. ACCT.2
    Log archive accountReceives logs from every account. Few people can read it; nobody can delete from it.
  3. ACCT.3
    Security tooling accountWhere alerts and posture findings are reviewed, away from the workloads.
  4. ACCT.4
    Non-PHI accountsDevelopment, marketing and experiments, seeded only with synthetic data.
  5. ACCT.5
    Organisation-level guardrailsNew in 2022Policies set from the top, for example limiting regions or stopping logging from being switched off.

In AWS Organizations, service control policies can cap what any role in a member account may do, which turns a written rule such as approved regions only into one that the platform enforces.

05 · Keys and logs

Encryption defaults, key management and logging

Defaults help. Key policy and log integrity are still your design.

Several AWS storage services now encrypt new data at rest by default with keys AWS manages. That covers the stolen-disk case. It does not separate who can read the data from who can administer it. For ePHI, most teams move to customer managed keys in AWS Key Management Service (KMS).

Key policy
States which roles may use a key and which may manage it. Keep the two lists apart.
Key per data class
Separate keys for the booking database, documents and backups, so one revocation does not stop everything.
Organisation trail
An AWS CloudTrail trail for the whole organisation, delivered to the log archive account.
Data-event logging
Control-plane logs show who changed settings. Reads of patient files need data events switched on where you need them.

06 · Checklist

A configuration checklist for the ePHI account

What a reviewer will ask to see, in one place.

  • Addendum on record. Accepted in Artifact, with the date and the accounts it covers.
  • Eligible-service check. Each service holding PHI matched to the current list, with the date checked.
  • Root user sealed. Hardware MFA, no access keys, use alerts.
  • Federated sign-in. People sign in through your identity provider with MFA; no long-lived user keys.
  • No public data. Public access blocked on storage, and no public snapshots or database endpoints.
  • Managed keys. Customer managed keys on ePHI stores and backups, with separate admin and use roles.
  • Logs centralised. Organisation trail to the log archive, protected from deletion, reviewed by a named person.
  • Restore tested. A recent, timed restore of the booking database, with the result recorded.

The shared-responsibility model itself, and what applies on any platform, is on the general HIPAA in the cloud page.

Sources: 45 CFR 160.103 and 45 CFR Part 164, Subpart C (Security Rule), in particular 164.308(b), 164.312(a) and 164.312(b); HHS Office for Civil Rights, Guidance on HIPAA and Cloud Computing; AWS public HIPAA compliance page; AWS Artifact user guide on managing agreements.

Questions about HIPAA on AWS

Does accepting the AWS addendum make our product HIPAA compliant?

No. It makes AWS your subcontractor for the eligible services.

The addendum covers what AWS does. Your access control, encryption choices, logging, staff training and other vendors remain yours, and AWS states there is no HIPAA certification for a cloud provider.

Can we use a service that is not on the eligible list?

Yes, as long as no PHI ever reaches it.

Ineligible services may run in the same account. The risk is accidental flow: a patient name in a log line or a tag. A separate ePHI account makes that far easier to prevent.

Should we accept at the account or the organisation level?

Organisation level avoids gaps; it also widens what counts as covered.

Organisation acceptance covers new accounts automatically. Pair it with guardrails so only the accounts meant for ePHI actually receive it.

Our customers ask for a BAA. Is that the AWS addendum?

No. Your customers sign a BAA with you.

Your clinic customers are covered entities and you are their business associate. You sign your own BAA with them, and AWS sits below you as a subcontractor under its addendum.

Check your AWS set-up before a customer does.

A HIPAA risk assessment reviews your AWS accounts, keys and logs against the Security Rule. Counsel should confirm how the addendum applies to your contracts.