GDPR, HIPAA and SOC 2: What Cloud Compliance Requires
Three frameworks, three different questions, one shared set of cloud controls. This guide explains what each asks of a deployment and where their requirements overlap in practice.
Compliance conversations usually start in the wrong place, with someone asking whether a provider is compliant. The question does not quite work. A platform can supply controls and documentation, but the obligation attaches to what you build on it and how you operate it, and nothing a provider does relieves you of that.
This guide describes, in general terms, what three commonly encountered frameworks expect of a cloud deployment. It is written for engineers planning infrastructure. It is not legal advice, the frameworks are interpreted differently across jurisdictions and sectors, and anything with real consequence should go past a qualified adviser and your own auditor before you rely on it.
The obligation follows the data, not the platform
Every one of these frameworks is concerned with information: personal data, health information, or the commitments you have made to customers about how you handle their records. The platform is relevant only insofar as it affects those commitments.
In practice this means a deployment on a heavily audited platform can still fall short, because the storage bucket was public, access was never reviewed, or nobody could produce a record of who did what. Conversely a deployment on a modest provider can satisfy an auditor if the controls are real and the evidence exists. The platform narrows what you have to build yourself; it does not decide the outcome.
Where the provider’s responsibility stops
Every provider publishes a responsibility split, and the boundary is consistent even when the wording differs. They operate the physical facilities, the hardware, the hypervisor and the managed services they run. You are responsible for your data, your identity configuration, your network rules, your operating systems and application code, and your encryption choices where you control them.
The line shifts with the service model. Rent a virtual machine and operating system patching is yours. Use a managed database and the provider handles the engine while you still own access control and the data inside it. The further up the stack you go, the more the provider covers, and the more carefully you need to read what remains with you.
A written agreement normally governs this. Under data protection law a processing agreement is typically required between you and any provider handling personal data on your behalf, and in healthcare contexts a comparable arrangement is expected before regulated information goes anywhere near the platform. Confirm the specific document exists and is signed before deployment, not during an assessment.
What data protection law asks of a deployment
GDPR is principle-based rather than prescriptive about technology, but its principles translate into concrete infrastructure decisions.
Location matters. You need to know which regions hold personal data, including backups, replicas and logs, and transfers outside the originating jurisdiction need a lawful basis. This makes region selection a compliance decision rather than a latency one, and it is worth reading European cloud providers and data sovereignty alongside how cloud regions and availability zones work before you place anything.
Minimisation is a design constraint. Collect and retain only what you need. In infrastructure terms this means retention policies that actually delete, and log fields that do not carry personal data without a reason.
Individual rights need to be technically possible. People can request a copy of their data or its deletion. If your architecture cannot locate every copy of one person’s records, including the ones in backups and analytics stores, you cannot satisfy the request. This is an architectural requirement that is very expensive to retrofit.
Breach notification runs on a clock. Short statutory deadlines mean detection and investigation capability is part of compliance, not separate from it.
What health information rules ask of a deployment
HIPAA applies to a defined set of entities handling protected health information in the United States, and it is more prescriptive about safeguards.
The expectations that land on infrastructure are access control limiting who can reach health information with unique identification per user, audit controls recording access to that information, integrity controls so that records cannot be altered undetected, transmission security so data is protected in transit, and encryption of stored data as an addressable safeguard that in practice is expected.
Two points catch teams out. Unique user identification rules out shared logins entirely, which is a stronger requirement than many organisations operate under by default. And audit controls apply to data access, not only to administrative actions, which means the data plane logging discussed in audit logging and activity monitoring from day one moves from optional to necessary on any store holding health information.
A formal agreement with any vendor that handles the information on your behalf is a precondition, and that typically includes the cloud provider and any subprocessor in the path.
What a service organisation audit asks of a deployment
SOC 2 differs from the other two in kind. It is not a law but an auditing framework, and the report describes whether the controls you said you have are designed appropriately and, in the more demanding form, whether they operated over a period of time.
That last clause is the whole character of it. You are not judged against a fixed technical checklist; you are judged against your own stated controls, consistently applied over months. A control that works on the day of the audit and was ignored for the preceding quarter is a finding.
The trust criteria cover security as a baseline, with availability, processing integrity, confidentiality and privacy added depending on scope. For infrastructure teams the recurring themes are access provisioning and removal with evidence, change management with review before deployment, monitoring with documented response, vendor management, and backup and recovery that has been tested rather than assumed.
Where the three overlap
Teams pursuing more than one framework often find the engineering work is largely shared. The controls below satisfy requirements in all three.
| Control | Why each framework cares |
|---|---|
| Unique identity per person, no shared logins | Attribution is the basis of every access requirement |
| Least privilege with periodic review | Limits exposure and produces review evidence |
| Multi-factor authentication on administrative access | Treated as a baseline expectation across all three |
| Encryption in transit and at rest | Named explicitly or implied by the protection standard |
| Audit logging with protected retention | Required for investigation, notification and evidence |
| Documented and tested backup and recovery | Availability and integrity obligations |
| Change management with review | Demonstrates controls are deliberate rather than accidental |
| Environment separation | Keeps regulated data out of lower-trust systems |
Build these once and most of the work transfers. The differences between frameworks then sit mostly in documentation, scope definition and the specifics of data location.
Evidence is the part that gets underestimated
Engineers tend to focus on implementing controls and treat the paperwork as an afterthought. Assessments run the other way around: an auditor cannot credit a control they cannot see demonstrated.
Plan for evidence at the same time as the control. That means access reviews that produce a dated document with names on it rather than a conversation. Change records that show who approved what and when. Alert history showing that something fired and someone acted. Recovery tests with a written result, including the time it took. A current inventory of systems and data flows, because scope cannot be assessed without one.
Generating this at the end of a period is nearly impossible and visibly reconstructed when attempted. Generating it continuously is a modest ongoing cost. Infrastructure defined as code helps considerably here, since the change history is the evidence, which is one of the less obvious arguments in the infrastructure as code provider model.
Choosing a provider with compliance in mind
Rather than asking whether a provider is compliant, ask a narrower set of questions you can verify yourself. Which regions can you pin data to, and does that include backups and logs? What agreements will the provider sign, and what is the process for obtaining them? What audit documentation do they make available to customers under what terms? Can you restrict administrative access and prove the restriction? What is their subprocessor list and how are changes notified?
Verify the answers in the provider’s own current documentation and contracts. Compliance posture changes over time and varies by region and by service, so a claim made anywhere other than the provider’s own materials should not be relied on, including claims in a marketplace listing. We are an independent marketplace and do not speak for any provider’s certification status.
Data location requirements frequently narrow the field first, which is why teams with European obligations often start from the Scaleway vs OVHcloud comparison and from listings such as Buy Scaleway Account or Buy OVH Account. Whichever you choose, the controls in the overlap table are yours to build, and they are what an assessment will actually examine.