The HIPAA Security Rule does not ask whether your cloud is well built. It asks whether specific safeguards for electronic protected health information were in place and operating. Those safeguards split three ways: some are policies you write, some belong to your cloud provider, and some are configuration that either exists in your accounts or does not.
That split is the whole planning problem. Teams either assume the cloud provider covers more than it does, or they try to answer administrative requirements with infrastructure and end up with neither. The useful question is narrow: which safeguards land on the foundation, and what does the foundation produce to prove them?
Which safeguards actually land on infrastructure?
#The Security Rule has three families. They do not divide evenly.
Administrative safeguards are the largest family by volume and the one infrastructure cannot touch. Risk analysis, a sanctions policy, workforce clearance, security awareness training, a contingency plan with testing, and business associate agreements down the chain: all of that is organizational work, and any vendor implying otherwise is selling something that does not exist.
Physical safeguards fall on the cloud provider under the shared responsibility model. You do not evidence them; you collect the provider’s reports and file them. In the AWS Enterprise Baseline, both auditor permission sets include the AWS Artifact report actions, so an auditor downloads those reports themselves instead of asking an engineer to do it.
Technical safeguards are the buildable ones, and they are where a foundation earns its place.
What each technical safeguard needs, concretely
#Here is the mapping that matters, with the mechanism rather than the intention. The controls in the right column describe the AWS Enterprise Baseline specifically.
| Technical safeguard | What it asks for | The mechanism in the AWS Enterprise Baseline |
|---|---|---|
| Access control, 164.312(a) | Unique user identification, emergency access, automatic logoff, encryption of stored data | IAM Identity Center as the only human access path, no IAM users anywhere, MFA enforced by a service control policy, 12 permission sets with production read-only for everyone except platform and DevOps leads, 12-hour sessions |
| Audit controls, 164.312(b) | Record and examine activity in systems that hold protected health information | Organization-wide multi-region CloudTrail, AWS Config recording in all 15 accounts with an organization aggregator, VPC Flow Logs on all six VPCs, GuardDuty in every account, seven CIS CloudWatch alarms |
| Integrity, 164.312(c) | Protect records from improper alteration or destruction | CloudTrail log file validation, versioning on the audit archive, and an audit-protection service control policy that denies stopping or deleting CloudTrail and Config anywhere in the organization |
| Authentication, 164.312(d) | Verify that a person or entity seeking access is who they claim to be | Single sign-on with MFA, federated deploy identities through GitHub OIDC rather than stored keys, and an unused-access analyzer that flags dormant access at 90 days |
| Transmission security, 164.312(e) | Guard against unauthorized access to information in transit | TLS-only bucket policies with a TLS 1.2 minimum, HTTPS-only load balancer listeners, and centralized egress through a hub VPC rather than a per-account internet path |
Two things are worth noticing. First, none of these controls are HIPAA-specific. They are the same controls a SOC 2 audit tests, which is why the evidence sources overlap almost completely. Second, every row is a configuration fact, which means every row can be checked continuously rather than asserted annually.
Required, addressable, and the word that gets people in trouble
#The Security Rule marks each implementation specification as either required or addressable. Encryption at rest is addressable. So is automatic logoff. So is the mechanism to authenticate electronic protected health information.
Addressable does not mean optional. It means: implement it if it is reasonable and appropriate, and if it is not, document why and implement an equivalent alternative measure. That second path is legitimate and almost always a bad trade. Writing a defensible memo about why you are not encrypting a database costs more effort than encrypting the database, and it converts a configuration fact into an argument you have to win twice — once with your auditor, once with the enterprise security team reading your questionnaire answers.
What the 130-rule HIPAA conformance pack checks
#AWS publishes a HIPAA Security conformance pack: a set of AWS Config rules mapped to the Security Rule’s expectations. The AWS Enterprise Baseline ships it as one of 29 conformance pack templates, at 130 rules, alongside a 37-rule encryption and keys pack that adds key-management checks such as rotation and deletion protection.
Both ship off, behind a single flag each, for a specific reason: Config rules are billed per evaluation, and the right set depends on which frameworks you are actually in scope for. Each switch states its price next to it, so the choice is made with the cost visible rather than discovered on a bill. The SOC 2 baseline pack is the exception, on by default at 13 rules mapped to Trust Services criteria, because it evaluates controls every customer wants regardless of framework.
There is no HIPAA standard in Security Hub. The pairing that works is the HIPAA pack for framework-mapped scoring plus AWS Foundational Security Best Practices for control-level findings across more than 300 controls. Between them you get a score per account and a finding per resource, which are the two shapes of question an assessor asks.
Then the honest caveat, which applies to every conformance pack in every cloud: a pack scores the resources you have, not the decisions you made. It will tell you that every EBS volume is encrypted. It cannot tell you that a support tool is writing patient identifiers into an application log, or that a dataset was exported to a laptop. Those are the incidents that actually get reported.
Where the business associate agreement line falls
#A business associate agreement is a contract. It establishes that a vendor handling protected health information on your behalf is bound to safeguard it and to notify you of breaches. Three boundaries are worth stating plainly, because they are where teams get surprised.
- The cloud provider’s agreement covers the services it names as eligible. Using a service outside that list to store, process or transmit protected health information puts you outside the agreement, whatever the architecture diagram says.
- The agreement configures nothing. Signing it does not encrypt a bucket, restrict a role or enable a log. It changes your legal posture and nothing about your technical posture.
- The chain runs both ways. Your customers will want an agreement from you, and every subprocessor touching protected health information needs one from you. The subprocessor list is itself a compliance artifact, which is one reason to keep a reusable trust pack rather than rebuilding the answer every quarter.
A useful discipline: keep an explicit inventory of which accounts and which services are allowed to hold protected health information, and make the account boundary the enforcement point. Account isolation is the strongest boundary a cloud provider offers, and it turns “where does this data live” into a list of accounts rather than a survey of tags. The same reasoning drives scope reduction under PCI DSS.
What readiness means here, and what it does not
#HIPAA has no certification. There is no body that issues a HIPAA certificate, and any badge you see is a vendor attesting to their own opinion. What exists is enforcement: investigations, usually triggered by a breach report or a complaint, in which a regulator asks for evidence that a safeguard was in place at a particular time.
That changes the shape of the work. An annual audit has a deadline you can sprint toward. An investigation arrives when something has already gone wrong, and it asks about a date in the past. The only way to answer is to have been recording continuously — which is why the recorders matter more than the packs, and why switching them on is the first thing to do rather than the last.
So: readiness means the technical safeguards are implemented, enforced above the account rather than inside it, and producing records you can query for any date in the period. It does not mean compliant, it does not mean certified, and it does not touch the administrative half of the rule. That line is the same one we hold for every framework on the compliance readiness page: controls and evidence are engineering, the opinion belongs to your auditor, and the risk analysis is yours.
A sequence that works
#If you are heading into a healthcare deal or a first workload in scope, this order produces the most defensible position per week of effort.
- Decide where protected health information is allowed to live, at account granularity, and write it down. Everything after this is easier when the boundary is an account rather than an intention.
- Turn on the recorders — audit logging, configuration recording, threat detection — because their history is what an investigation asks for, and history cannot be backdated.
- Set the encryption defaults at the account level, with customer-managed keys and rotation, so the addressable specification is satisfied by default rather than by review.
- Remove standing human access to production. Read-only for everyone except a named short list, with elevation that is logged.
- Switch on the HIPAA pack and the encryption and keys pack at least two weeks before anyone needs to read the score. An organization-wide pack takes time to reach every account, and the first evaluation needs to settle.
- Rehearse the export. Pull a quarter of evidence as though a regulator asked. Anything that takes more than an hour is a gap you have not found yet. The mechanics of that rehearsal are in evidence automation.
Steps one through four are worth doing whether or not HIPAA ever applies to you, which is usually the sign that they are the right steps.
What we do not claim
#We do not sign a business associate agreement on your behalf, we do not decide whether a workload handles protected health information, and we do not write your risk analysis, your policies or your training. No foundation makes an organization compliant, and no conformance pack score is a legal position.
What the AWS Enterprise Baseline does deliver is the technical safeguard half, built in rather than retrofitted: the identity model, the guardrails, a log archive that cannot be stopped from inside the accounts it records, encryption defaults set per account, and 11 frameworks with documented control mappings so that “show me where this requirement is met” is a lookup instead of an investigation.
If you are about to sign a healthcare customer and want to know which of those safeguards you already have, the BuiltForProd Assessment measures an existing environment against the BuiltForProd Standard and names the gaps — including the ones that are much cheaper to fix before a contract obliges you to.