HIPAA compliance for startups is less about paperwork than founders fear and more about architecture than they expect. If you build software that stores, processes or transmits protected health information (PHI) on behalf of a healthcare provider, health plan or clearinghouse, you are almost certainly a business associate under HIPAA. That means signed business associate agreements (BAAs), a written risk analysis, and a system that can prove who accessed which record and when. The good news: in an MVP, the technical part costs a few extra days if you design for it in week one. Retrofitting it after launch can cost you a pilot.
I'm Kunal Vohra. I've been the technical co-founder six times, I built SYNCHFIT (an AI fitness and health product), and my M.Tech is in cybersecurity, so access control and data protection are the discipline I trained in. I'm not a lawyer or a compliance auditor, and nothing here is legal advice. This is the engineering side: what I build into a healthtech MVP from the first commit, and what I deliberately leave for later.
Does HIPAA apply to your startup?
HIPAA applies to your startup if you are a covered entity (a healthcare provider that bills electronically, a health plan, or a clearinghouse) or a business associate of one, meaning you handle PHI on their behalf. Most healthtech startups selling to clinics, hospitals, payers or digital health companies are business associates. Many direct-to-consumer health apps are not covered by HIPAA at all, but that does not mean they are unregulated.
Three questions settle it quickly:
- Who is your customer? If a provider, payer or another business associate pays you and their patient data flows through your product, you are a business associate. You will sign a BAA with them before the pilot starts, and you need BAAs with your own vendors that touch that data.
- Where does the data come from? If a consumer types their own symptoms, steps or sleep into your app and no covered entity is involved, HIPAA usually does not apply. The FTC's Health Breach Notification Rule and state health-privacy laws (Washington's My Health My Data Act is the one founders hear about most) often do.
- Where are you going next? A consumer app that plans to sell to health plans or integrate with an EHR will cross into HIPAA territory. Build for where the company is going, because the architecture is the expensive part to change.
Users outside the US bring GDPR or, in India, the DPDP Act (see my DPDP Act explainer).
What counts as PHI (and what does not)
PHI is any individually identifiable health information: health data linked to anything that can identify the person, such as a name, email, phone number, address, dates more specific than the year, device identifiers or IP addresses. On its own, a blood pressure reading is just a number. The same reading next to an email address is PHI.
This matters for engineering because PHI is rarely confined to the obvious tables. It shows up in:
- Logs and error trackers, when a request body or a stack trace includes a patient record.
- Analytics and session replay, when a tool records what a user typed into a symptom form.
- Marketing pixels on patient-facing pages, which can send identifiable health information to an ad platform. It is one of the most commonly reported failures in the sector, usually introduced by the marketing stack.
- LLM prompts and prompt logs, the newest leak in the chain.
Properly de-identified data (HIPAA's Safe Harbor method removes eighteen listed identifiers; the alternative is an expert determination) is no longer PHI, which is often the cleanest way to keep an analytics or evaluation tool out of scope.
The PHI boundary: the most important HIPAA decision in your MVP
The single most valuable thing you can do for HIPAA compliance at MVP stage is to draw a PHI boundary before a screen is designed: a short document and diagram listing which fields are PHI and which systems are allowed to see them. Everything outside the boundary never receives PHI. Everything inside it has a BAA, access control and audit logging.
Drawing the PHI boundary takes an afternoon at MVP stage. Drawing it after you have live patient data in six vendor systems is a migration project, and it usually happens in the middle of a security review.
That diagram decides your analytics tool, error tracker, logging, email provider and whether an LLM can touch the data at all. It also puts HIPAA's minimum necessary principle into practice. I do this in the scoping week of every healthtech build. My guide to scoping an MVP covers the rest of that week.
The cost of skipping it is concrete. On one engagement, a signed enterprise pilot stalled in security review because PHI was flowing into three vendor systems with no BAA and no audit trail. Redrawing the boundary, removing PHI from everything that did not need it and adding immutable access logging took six weeks, and the deal closed. The same work after a failed audit would have cost the deal.
BAAs: every vendor that touches PHI needs one
A business associate agreement is the contract in which a vendor commits to protect PHI the way HIPAA requires. You sign one with your customer, and you need one with every vendor of yours that creates, receives, stores or transmits PHI. A vendor that will not sign a BAA cannot be inside your PHI boundary, however good the tool is.
How this plays out in a typical healthtech stack:
- Cloud hosting, database and storage. The major cloud providers sign BAAs, but only for their HIPAA-eligible services. Check each service you use against the list and turn on the configuration it requires. A separate managed database vendor needs its own BAA.
- Email, SMS and notifications. Many providers will not sign a BAA, or only on enterprise plans. Keep PHI out of message bodies anyway: "You have a new message in your portal", not the lab result.
- Error tracking, logging and analytics. The most common gap. Either pick a vendor that signs, or scrub PHI before it leaves your servers and keep the tool outside the boundary.
- LLM providers. Some model providers offer BAAs on specific API or enterprise tiers, with zero or limited data retention; others do not offer one at all, and consumer chat products almost never qualify. I check this at vendor-selection time, because discovering it late means redesigning the feature. My AI API setup guide covers how to classify data before choosing a provider.
The technical safeguards to build into a healthtech MVP from day one
The HIPAA Security Rule groups its requirements into administrative, physical and technical safeguards. For a cloud-hosted startup, physical safeguards are mostly inherited from your cloud provider, and the technical ones are what your engineering team builds. These are the ones I treat as non-negotiable in v1:
1. Real access control with role separation. Patient, clinician, admin and support are different roles, enforced on the server, not hidden in the UI. Unique logins; no shared accounts.
2. Multi-factor authentication for every staff and admin account, and for your cloud, code and vendor consoles.
3. Encryption in transit and at rest. TLS everywhere, encrypted databases, storage and backups. Under the current Security Rule, encryption is technically "addressable" rather than strictly required. HHS proposed an update in early 2025 that would make encryption and MFA explicit requirements; whatever its status when you read this, build as though it is already law. In a modern cloud stack it costs almost nothing.
4. Append-only audit logging on every PHI read and write. HIPAA expects you to be able to show who accessed what. Every health system that buys from you will ask the same question in its security questionnaire. Build logs that answer "who saw this patient's record, and when", not "which service emitted a line". In an MVP this is close to free because the data model is still forming. On a live system it means backfilling history you do not have.
5. Session timeouts and backups you have actually restored. Clinician screens sit on shared workstations, and availability is part of HIPAA, not just confidentiality.
6. Secrets out of code and least-privilege service accounts. The same basics as any startup, and the ones I fix first in my startup security checklist.
None of these adds screens. In my builds the audit logging and access control show up as a separate backend line in the scope, a few days of work, rather than being folded into the per-screen rate. If you are budgeting, the MVP cost calculator includes a security review option, and my MVP timeline post explains why compliance requirements need to be named in week zero.
The paperwork part: risk analysis, policies and breach notification
HIPAA compliance is broader than software. Alongside the technical safeguards, a business associate needs:
- A written risk analysis: what PHI you hold, where, what could go wrong, and what you are doing about it. It is required, it is the first document an auditor or a customer's security team asks for, and the PHI boundary diagram is most of its first draft.
- Policies and workforce training: who may access PHI and how access is granted and revoked. Short and true beats long and copied.
- A breach and incident plan. HIPAA's Breach Notification Rule sets deadlines (individuals must be told without unreasonable delay and no later than 60 days after discovery, and as a business associate you must notify your covered-entity customer). The time to decide who leads an incident is before you have one.
- A designated security owner, even if it is a co-founder wearing a second hat.
There is no official HIPAA certification; HHS does not certify products or vendors. Buyers want to see the risk analysis, your BAAs, your policies and, later, a SOC 2 or HITRUST report.
What a healthtech MVP can leave for later
A HIPAA-aware MVP is not an enterprise compliance programme. These can usually wait until a signed customer needs them:
- SOC 2 and HITRUST. Providers, payers and large employers will ask, but these are sales requirements that arrive with enterprise deals. If the architecture is right, most controls are the same basics with evidence attached.
- Full FHIR or HL7 interoperability. Scope the narrow slice your first pilot requires, usually a handful of resource types, and structure it so the rest extends without a rewrite.
- Formal penetration testing before real users (a code security review before the first pilot is still worth it).
- Patient messaging, scheduling and native mobile, unless they are the product.
What cannot wait: the PHI boundary, BAAs with in-scope vendors, access control, MFA, encryption and audit logging. Those are the decisions that get more expensive every week you postpone them.
HIPAA and AI features: the newest compliance gap
AI features are the fastest-growing HIPAA gap, because PHI reaches a model provider through a prompt, a retrieval step or a prompt log before anyone checks the vendor's terms. Before any model touches patient data, you need a signed BAA with that provider (on a tier that offers one), clear retention terms, and PHI kept out of prompt logs and evaluation sets unless they are inside the boundary.
If a model output influences care, design the failure mode first: who reviews the output before it reaches a clinician or patient, what gets logged, and how a clinician overrides it. Agents that call tools add another layer, covered in my AI agent security guide, and investors ask about all of it in technical due diligence.
What HIPAA compliance costs a startup, and who should own it
For an MVP, building the technical safeguards in from the start adds days, not months. The larger costs (formal audits, SOC 2 or HITRUST, specialist legal help) come later, usually with enterprise customers. Retrofitting the architecture after launch is where startups lose real money and time.
The owner should be whoever owns your architecture, usually the CTO. For a pre-seed or seed healthtech company, that rarely needs to be a full-time CTO at $180,000 to $260,000 a year plus equity. A technical advisor retainer ($750 a month for 15+ hours, held through December 2026) covers a review of your PHI boundary, vendor list and logging. If you need someone to own the architecture, the security reviews and the hires week to week, that is a fractional CTO for healthcare engagement ($2,500 a month for 50+ hours on the same terms); the fractional CTO cost guide puts market rates beside mine. If you have not built yet, my team and I build healthtech MVPs this way by default, with screens from $30 and a typical v1 of eight to fourteen screens.
One line to be clear about: I'm not a HIPAA compliance consultant. For audits and legal sign-off, use a specialist firm; my job is making sure the system they audit is built correctly.
Frequently asked questions
Does HIPAA apply to my health app startup? It applies if you are a covered entity or, far more often for startups, a business associate handling PHI on behalf of a provider, health plan or clearinghouse. A consumer app where users enter their own data with no covered entity involved is usually outside HIPAA, though the FTC Health Breach Notification Rule and state health-privacy laws may still apply.
Is there a HIPAA certification for software? No. HHS does not certify products or companies as HIPAA compliant. Buyers look at your risk analysis, your BAAs, your policies and your architecture, and later at a SOC 2 or HITRUST report.
Do I need a BAA with my cloud provider and my LLM provider? Yes, with every vendor that creates, receives, stores or transmits PHI for you. Major cloud providers sign BAAs covering their HIPAA-eligible services. Some LLM providers offer BAAs on specific API or enterprise tiers; if yours does not, PHI must not reach it.
How much does HIPAA compliance cost for an MVP? Built in from the start, the technical part (PHI boundary, access control, MFA, encryption, audit logging) adds days of backend work, not a separate project. Audits, SOC 2 or HITRUST and legal review are later, larger costs that usually arrive with enterprise customers.
Can a HIPAA-compliant MVP use AI? Yes, if the model provider signs a BAA on the tier you use, retention terms are clear, and PHI stays out of prompt logs and test sets outside your boundary. If a model influences care, design the human review and the failure mode before the model choice.
If you're building a healthtech product and you're not sure where PHI is flowing today, book a 30-minute call. Bring your vendor list and a rough diagram of your data, and I'll tell you where your PHI boundary should sit, which vendors are a problem, and what to fix before your first security review. Or get in touch if you'd rather write first.
Written By
Kunal Vohra
Technical Co-Founder & Fractional CTO
I've co-founded 6+ startups across India, the UAE, and the US, spanning AI, Web3, fintech, and cybersecurity. I write about the technical and strategic decisions that determine whether a startup thrives or stalls.
Comments
Loading comments…