Quick answer: Software as a medical device (SaMD) is software with a medical purpose that runs on its own, without being part of a hardware device. The FDA regulates it by risk: you work out whether it is a device at all, set your SaMD FDA classification (Class I, II or III), pick a pathway (510(k), De Novo or PMA), and build a quality system and cybersecurity file alongside the code. In 2026, the biggest changes are looser rules for some decision-support and wellness software, plus the new quality regulation (QMSR) that took effect on February 2.
You built something clever. Maybe it reads retinal scans, scores sepsis risk or tells a diabetic how much insulin to take. Then someone on your advisory board casually asks, “Have you thought about the FDA?” and your stomach drops into your shoes. Welcome to software as a medical device, the corner of healthcare where shipping on Friday is not a thing.
Here is the good news: the rules are learnable, and they follow a logic any engineer will recognize. Think of the FDA as the strictest code reviewer you have ever met, one that wants to see your requirements, your tests, your risk analysis and your release process before it lets you merge. If you are a MedTech founder, a HealthTech CEO or a digital health product leader, this guide gives you the order of operations. And when you reach the build stage, our mobile app development team can show you what a testable, documented build looks like.
By the end you will know how to tell whether your product counts as software as a medical device, how SaMD FDA classification really works, which 2026 changes matter, a seven-step roadmap from idea to clearance, the mistakes that burn founders, and how to pick medical device software development partners without getting burned.
Is your product actually software as a medical device?
Your product is software as a medical device when its intended use is a medical purpose, like diagnosing, treating, monitoring or preventing disease, and it does that job without being part of a hardware device. The IMDRF wrote that definition, and the FDA applies it, so it is the one that counts.
Definition: intended use. What you claim your software does, in your labeling, your website, your sales deck and even your investor pitch. The FDA looks at all of it. Change the claim and you can change the regulation, which is why intended use is the first decision you make, not the last.
Not everything health-related is a device, and 2026 shifted the line. In January, the FDA updated its guidance on clinical decision support software and general wellness products. Summaries from law firms say the FDA now takes a lighter touch with some software that gives a single clinically appropriate recommendation and with certain non-invasive wearables, as covered by Akin and Ropes & Gray. The same coverage notes that time-critical alerts, such as stroke or sepsis detection, still look like device software.
| Example | Likely status | Why |
|---|---|---|
| App that reads chest X-rays and flags pneumothorax | Device (SaMD) | Analyzes medical images for a diagnostic purpose |
| Dose calculator that tells a patient how much insulin to take | Device (SaMD) | Directly drives treatment |
| Sepsis or stroke alert for clinicians | Device (SaMD) | Time-critical decision support |
| Software that displays guideline-based options for a clinician to review | May be non-device CDS | Clinician can review the basis independently |
| Step counter and sleep tracker for general fitness | Often general wellness | Low risk, no disease claims |
| Appointment scheduling for a clinic | Not a device | Administrative function |
Key takeaway: Decide your intended use deliberately. It determines whether your software as a medical device needs the FDA at all.
How does SaMD FDA classification work?
SaMD FDA classification sorts your software into Class I, II or III based on the risk it poses to patients if it is wrong, and that class decides your pathway. Most software lands in Class II, where a 510(k) is the usual route.
| Class | Risk level | Typical pathway | Software example |
|---|---|---|---|
| Class I | Low | Often exempt from premarket review, general controls apply | Low-risk data display tools |
| Class II | Moderate | 510(k) with a predicate, or De Novo if none exists | AI imaging triage, many diagnostic aids |
| Class III | High | Premarket approval (PMA) | Software that drives life-sustaining decisions |
Three pathways matter. A 510(k) shows your software is substantially equivalent to a legally marketed predicate. A De Novo request is for novel, low-to-moderate risk products with no predicate. A PMA is the heavy route for high-risk devices and requires the most evidence.
The numbers tell you which road is busiest. The FDA’s AI-enabled device list passed 1,600 authorizations by September 2026, and an analysis of that list found that 96% went through the 510(k) pathway, with radiology accounting for 76%. Translation: the 510(k) road is well-paved, but if you are building another radiology algorithm you will have lots of company.
Behind the class sits a second lens. The IMDRF framework scores software by how serious the healthcare situation is and how significant the information is to the decision, and regulators use that thinking to judge how much evidence you need. A tool that informs a routine decision needs less proof than one that drives treatment in a critical condition.
Key takeaway: Your class sets your pathway, and your pathway sets your budget and timeline. Get the SaMD FDA classification conversation going early.
What are the new FDA regulations for software as a medical device in 2026?
Five changes matter most for software as a medical device in 2026: relaxed rules for some decision-support and wellness software, a withdrawn clinical evaluation guidance, the new QMSR, updated cybersecurity expectations and the AI change-control framework now in use.
| When | What changed | What it means for you |
|---|---|---|
| Jan 2026 | Updated CDS and general wellness guidance | Some software may now fall outside device rules, but only if it meets every criterion |
| Jan 2026 | SaMD Clinical Evaluation guidance withdrawn | Do not lean on it; discuss evidence plans with the FDA directly |
| Feb 2, 2026 | QMSR took effect | Your quality system must align with ISO 13485:2016 concepts |
| Feb 3, 2026 | Cybersecurity premarket guidance reissued | SBOM, threat model and vulnerability plan belong in your submission |
| Ongoing | PCCP guidance for AI-enabled software | Pre-agree on model updates so retraining does not mean resubmitting |
A few details are worth knowing. The FDA withdrew its SaMD Clinical Evaluation guidance in January, which means founders should plan evidence with the FDA through a pre-submission meeting rather than relying on that document. The Quality Management System Regulation replaced the old Quality System Regulation on February 2 and pulled in ISO 13485:2016 by reference. And the FDA’s cybersecurity guidance, issued February 3, 2026, supersedes the June 2025 version and covers the SBOM and vulnerability management expectations for cyber devices.
If your software learns, read the guidance on predetermined change control plans. A PCCP lets you agree with the FDA in advance on which model updates you will ship and how you will test them, so you are not resubmitting every time you retrain.
One caution: guidance moves, and one law-firm summary notes the CDS guidance was reissued again later in January. Always pull the current version from the FDA before you make a classification call.
Key takeaway: 2026 loosened some doors and tightened others. Quality and cybersecurity expectations are firmer, while some decision-support software got more room.
What is the step-by-step roadmap to FDA clearance for software as a medical device?
The roadmap has seven steps, and you can run several in parallel once your intended use is locked. Here is the order most teams follow.
Step 1: Lock your intended use and indications
Write one sentence: who uses it, for what clinical purpose, in what setting. Everything else, from classification to your marketing copy, hangs off this sentence.
Step 2: Classify and pick a pathway
Search the FDA databases for similar cleared products, identify possible predicates and product codes, and decide between 510(k), De Novo and PMA. If you are unsure, a pre-submission meeting with the FDA is the cheapest way to find out.
Step 3: Stand up your quality system
Under the QMSR you need design controls, document control, risk management and corrective action. Most software teams use IEC 62304 for the software lifecycle and ISO 14971 for risk management, alongside ISO 13485 concepts. This is where “move fast” meets “show your work.”
Step 4: Build with traceability
Write requirements, link them to tests, link tests to risks. Reviewers love a traceability matrix the way engineers love green CI.
Step 5: Bake in cybersecurity
Create a threat model, keep a software bill of materials, plan for patching and vulnerability disclosure, and design authentication and encryption in from the start.
Step 6: Generate evidence
Run verification and validation, plus clinical or performance testing appropriate to your risk. Match the evidence to the claim: bigger claims need stronger proof.
Step 7: Submit, launch and monitor
Assemble the submission, respond to FDA questions, then run post-market surveillance: complaint handling, adverse event reporting, and controlled updates. Launch is the starting gun, not the finish line.
Key takeaway: Treat the roadmap like an architecture: decide intended use first, then let quality, security and evidence grow around it.
Where do founders trip up with software as a medical device?
Usually in the same five places, and all of them are avoidable.
- Marketing writes the claims. A single “detects cancer” line on your homepage can turn a wellness app into a regulated device.
- Quality system as an afterthought. Retrofitting design controls after the code is written costs far more than building them in.
- Skipping the pre-submission meeting. It is one of the cheapest ways to avoid a very expensive surprise.
- Treating cybersecurity as a checklist. Reviewers want a threat model and a postmarket plan, not a generic policy.
- Letting a vendor ship “AI-generated” code with no review trail. If you cannot show who reviewed what, you cannot show control.
Building Software That May Be a Medical Device?
EnactOn builds the software layer with senior-led architecture, a four-gate QA pipeline and human-reviewed AI acceleration, working alongside your regulatory and quality advisers.
How do you choose medical device software development partners?
Choose medical device software development partners who can show documentation habits, not only code. A great engineering team that cannot produce traceable requirements, tests and risk records will slow your submission down.
| Ask this | A good answer sounds like | A red flag sounds like |
|---|---|---|
| How do you document design and requirements? | Traceable requirements linked to tests and risks | ”We keep it in Slack and the repo” |
| How do you handle AI-generated code? | Specification-first, human review, no raw output in production | ”The model writes most of it” |
| What is your security process? | Threat modeling, SBOM and dependency scanning | A generic compliance PDF |
| Who owns the regulatory strategy? | You and your regulatory consultants, with engineering supporting | ”We handle FDA for you” with no named expert |
| How do you test? | Automated suites plus real-device testing | Manual spot checks before release |
Be wary of anyone who says they will “handle the FDA” alone. Regulatory strategy belongs with qualified regulatory and quality specialists, and a good build partner works with them. Our enterprise software development page shows how we scope integration-heavy products with that kind of split in mind.
Key takeaway: The right build partner makes your regulatory consultant’s life easier, not harder.
What should your first 90 days on a SaMD project look like?
Spend the first 90 days on decisions, not features: lock intended use, classify, book a pre-submission meeting and stand up a lightweight quality system, so the code you write from week four onward is already traceable.
| Window | Focus | Output |
|---|---|---|
| Days 1 to 30 | Intended use, classification research, predicate search | Written intended use and pathway hypothesis |
| Days 31 to 60 | Quality system setup, risk analysis, architecture and security approach | QMS procedures, risk file, threat model v1 |
| Days 61 to 90 | Pre-submission prep, requirements and test plan, first traceable build | Pre-sub package and traceability matrix |
Conclusion
Software as a medical device is regulated by risk, documented by habit and cleared by evidence. If you decide your intended use early, classify carefully, build quality and cybersecurity in from day one and plan your evidence around your claims, the FDA stops feeling like a wall and starts feeling like a process.
Your next step this week: write your one-sentence intended use, search for similar cleared products, and book time with a regulatory specialist to test your classification. Then line up engineering partners who can show traceable work, because your SaMD FDA classification will shape their job as much as yours.
EnactOn is a Clutch Certified engineering partner, and medical is one of the largest industries in our Clutch client mix. We build software as a medical device with senior-led architecture, human-reviewed AI acceleration and a four-gate QA pipeline, and we work alongside your regulatory advisers rather than replacing them.
FAQs
What are the FDA guidelines for software as a medical device (SaMD)?
The FDA does not have one software as a medical device (SaMD) rulebook. It applies the device framework to software through several pieces: the device definition and your intended use, classification (Class I, II or III) and the matching pathway, quality system requirements under the QMSR, cybersecurity expectations including an SBOM for cyber devices, and AI-specific guidance such as predetermined change control plans. It also publishes guidance on clinical decision support and general wellness software to define what falls outside device regulation. Start with the FDA’s SaMD page and use a pre-submission meeting to confirm how it applies to your product.
What are the new FDA regulations for 2026?
The headline changes are the January updates to clinical decision support and general wellness guidance, which give some software more room, and the withdrawal of the software as a medical device (SaMD) Clinical Evaluation guidance. The Quality Management System Regulation took effect on February 2, aligning US requirements with ISO 13485:2016, and a reissued premarket cybersecurity guidance followed on February 3. PCCP guidance for AI-enabled software also shapes how learning products are authorized. Always check the FDA site for the current version before making a decision.
How does SaMD FDA classification work?
The FDA sorts software into Class I, II or III based on the risk to patients if it fails. Class I is low risk and often exempt from premarket review, Class II is moderate risk and usually needs a 510(k) or De Novo, and Class III is high risk and requires premarket approval. Your intended use drives the class, so get that sentence right first.
What is the difference between a 510(k), De Novo and PMA for software?
A 510(k) shows your software is substantially equivalent to a legally marketed predicate. A De Novo request is for novel products with low to moderate risk and no predicate. A PMA is the most demanding route and is used for high-risk software that needs the strongest evidence. Most software as a medical device (SaMD) products use 510(k) or De Novo.
When is my health software not a medical device?
It can fall outside device rules if it is purely administrative, qualifies as general wellness, or meets all the criteria for non-device clinical decision support. The 2026 guidance widened the room for some of these, but time-critical alerts and software that analyzes images or signals generally remain devices. Because the criteria are detailed, document your reasoning and test it with a regulatory specialist.
Do I need ISO 13485 and a quality system as a startup?
If your software is a regulated device, you need a quality system that meets the QMSR, which incorporates ISO 13485:2016. That means design controls, document control, risk management and corrective action, even for a small team. Many startups use lightweight, software-friendly tooling, but the records must exist and be auditable.
What is a PCCP and do I need one for AI?
A predetermined change control plan describes the model changes you plan to make after clearance and how you will validate them. If your product retrains or updates its algorithm, a PCCP can let you ship pre-agreed changes without a new submission each time. It is not mandatory for every AI product, but it is worth discussing with the FDA early.
Does the FDA require cybersecurity documentation for software as a medical device (SaMD)?
Yes, for cyber devices. Section 524B of the FD&C Act requires a software bill of materials, a plan to monitor and address vulnerabilities, and evidence of reasonable assurance of cybersecurity in premarket submissions. The FDA’s latest cybersecurity guidance explains the expectations in detail. Build threat modeling and dependency tracking into development rather than assembling it at the end.
How long and how expensive is FDA clearance for software?
It varies a lot with class, pathway, evidence needs and how ready your quality system is. The FDA review itself is only one part of the timeline, and evidence generation and quality system setup often take longer than the coding. Get a scoped estimate from your regulatory consultant after you have written your intended use and chosen a pathway.
How do I choose medical device software development partners?
Look for traceable documentation, a clear security process, human review of any AI-generated code, and real-device testing. Ask who owns regulatory strategy; the answer should be you and your regulatory advisers, with the engineering team supporting. Reference calls with clients in regulated spaces are worth the time.
How does EnactOn help with software as a medical device projects?
We start with discovery, where our Target-Market Alignment work typically removes 30 to 40 percent of non-essential features, then build with senior-led architecture and a four-gate pipeline covering functional, security, performance and deployment checks. AI speeds up delivery under strict human review. We architect for HIPAA data handling and GDPR privacy-by-design, and we work alongside your regulatory consultants rather than providing regulatory submissions ourselves.






