“Make it compliant” hides the first architecture question: which regime applies to which data flow, entity and deployment? GDPR is a data-protection law, HIPAA governs regulated US healthcare actors and protected health information, and FedRAMP is a federal cloud authorization and ongoing-evidence program. They overlap on security controls, but their load-bearing constraints are not interchangeable.
Choose a preset or toggle controls manually. Each bar counts requirements represented in this teaching model; red blockers are regime-specific design decisions still missing. A generic “secure AI” baseline covers only 2/7 GDPR capabilities, 2/5 HIPAA capabilities and 2/7 FedRAMP capabilities.
| regime | scope question | load-bearing constraints | architecture consequence |
|---|---|---|---|
| GDPR | Are you processing personal data in a context within its territorial/material scope? Who is controller vs processor? | lawful basis and purpose limitation; minimization; processor terms; data-subject rights; retention/erasure with applicable exceptions; safeguards for relevant international transfers; privacy by design | inventory every copy and purpose; tag provenance/processing purpose; build access/correction/deletion workflows through source, index, cache, trace and backup policy; control processor/subprocessor and transfer paths |
| HIPAA | Is a covered entity or business associate creating, receiving, maintaining or transmitting PHI/ePHI? | permitted uses/disclosures and minimum necessary; BAAs with business associates/subcontractors where required; administrative, physical and technical safeguards; unique access and audit controls; incident/breach duties | draw the PHI boundary before the pilot; do not send PHI to a service until the contractual and safeguard path is approved; scope retrieval/tools to minimum necessary; retain protected, reviewable audit evidence |
| FedRAMP | Is a cloud service offering handling federal information for agency use, and what impact level/path does the agency require? | defined authorization boundary; NIST SP 800-53 baseline implementation and assessment evidence; external-service/inheritance story; agency authorization/acceptance; continuous monitoring, vulnerability and POA&M operations; required validated cryptography | choose dependencies from an acceptable authorized path; include model/API calls and metadata flows in the boundary analysis; maintain SSP/control evidence; design change control, scans, incident reporting and ConMon as product operations |
The preset measurements make the non-equivalence concrete. The privacy architecture reaches 7/7 on the GDPR teaching lens but remains 2/5 for HIPAA because it has no BAA, PHI boundary or minimum-necessary design; it remains 2/7 for FedRAMP because it has no authorization boundary, accepted dependency path, control evidence or continuous monitoring. The healthcare preset reaches 5/5 HIPAA and still does not acquire GDPR rights workflows or a FedRAMP authorization package by resemblance. Shared controls are reusable; scope and regime-specific evidence are not.
# the architect's first-page regime map data_flow -> people/data/agency context -> applicable regime(s) -> accountable role GDPR -> purpose + lawful basis + minimization + rights + transfers + deletion propagation HIPAA -> PHI boundary + BAA chain + minimum necessary + access/audit + incident process FedRAMP -> authorization boundary + accepted services + control evidence + continuous monitoring # shared control evidence can be reused; the applicability decision and package cannot owner = privacy_officer | security_officer | authorizing_official | counsel evidence = data_map + contracts + tests + logs + assessments + change/incident records
Primary sources (verify the current version with your reviewers): EUR-Lex · GDPR · HHS · HIPAA Security Rule summary · HHS · Business Associates · FedRAMP · CSP Authorization Playbook.