For connected medical device manufacturers, the cybersecurity part of an FDA submission has become one of the main sources of questions, and even of refusals to accept. The good news: the expectations are public, stable and largely predictable. A file that answers them point by point gets through review without surprises.
This article is designed to be reused internally: it describes what is actually assessed, proposes a method to audit your own file before submission, provides table templates to copy into your documents, and ends with a checklist. For the regulatory framework itself (section 524B, definition of a "cyber device"), see our page Medical device cybersecurity: FDA requirements.
Which "audit" are we talking about?
Manufacturers often talk about an "FDA audit", but the phrase covers three distinct moments, which involve different people and different evidence.
| Moment | Who assesses | What is examined |
|---|---|---|
| Submission review (510(k), De Novo, PMA) | FDA reviewers, including cybersecurity specialists | The cybersecurity documentation in the submission (eSTAR format for 510(k) and De Novo), against section 524B and the premarket guidance |
| Quality system inspection | FDA investigators, on site | Quality system records (QMSR, aligned with ISO 13485 since February 2026): design controls, risk management, complaints, corrective actions, including for cybersecurity |
| Independent assessment commissioned by the manufacturer | An external provider (gap analysis, penetration test) | The device itself and its file, to produce test evidence and close gaps before the FDA sees them |
The key point: the FDA reviewer does not test your device. They read your file. Anything that is not written, traced and justified does not exist for them. Preparation therefore means producing evidence that a third party, who knows neither your product nor your team, can follow.
The key points the FDA checks
The guidance Cybersecurity in Medical Devices: Quality System Considerations and Content of Premarket Submissions describes the expected content. The table below restates it from the reviewer's point of view, with the mistakes we most often find during our gap analyses. It can be used directly as a self-assessment grid.
| Item | What the reviewer checks | Common mistake |
|---|---|---|
| Threat model | The whole system is covered: device, mobile app, gateway, cloud, service and manufacturing tools | Only the device enclosure is modeled; cloud and service tool are forgotten |
| Cybersecurity risk assessment | Risks are assessed on exploitability, linked to patient safety risks (ISO 14971), and residual risk is justified | Reuse of the ISO 14971 analysis with probabilities of occurrence, which do not fit intentional threats |
| Architecture views | Global system view, multi-patient harm view, updatability/patchability view, security use case views | A single network diagram, with no flows and no trust boundaries |
| SBOM | All software components, with version, supplier, level of support and end-of-support date | Application dependencies only, without the OS, firmware, embedded libraries or end-of-support dates |
| Known vulnerabilities | Each known vulnerability in SBOM components is assessed, with a rationale and, where relevant, a mitigation | A raw scanner export with no case-by-case analysis |
| Security controls | Authentication, authorization, cryptography, code and data integrity, logging, resiliency, secure updates | Controls listed with no description of how they are implemented or verified |
| Traceability | Each threat leads to a control, each control to a test, each test to a result | Documents that are consistent individually but not linked together |
| Cybersecurity testing | Requirements verification, penetration testing, fuzzing, software composition analysis, by independent testers, with a detailed report | A two-page report concluding "no critical vulnerability" with no method or scope |
| Unresolved anomalies | Software defects known at submission time are assessed for security impact | The list is provided without any cybersecurity assessment |
| Labeling | Labeling describes controls, ports and interfaces, secure configuration, updates, end of support and SBOM availability | No mention of cybersecurity in the instructions for use |
| Postmarket management plan | Vulnerability monitoring, coordinated vulnerability disclosure (CVD), remediation timelines, customer communication | A generic policy with no timelines, no owner and no reporting channel |
Preparing your audit in six steps
Each step produces a deliverable that goes straight into the file. The order matters: tests are only valuable if they verify a threat model that already exists.
1. Scope the system and appoint an owner
Appoint a cybersecurity owner for the file, separate from the regulatory project lead but working closely with them. Then inventory the system as the FDA sees it: the device, but also everything that connects to it or updates it. List every interface (radio, USB, serial port, JTAG, Ethernet, cloud API) and state whether it is active, disabled in production or protected.
Deliverable: global system diagram and interface inventory.
2. Run a gap analysis against the FDA grid
Download the eSTAR template for your submission type and start filling in its cybersecurity section now, even partially. It is the exact grid the reviewer will read: every empty field is a gap. Combine it with the key points table above, recording for each line the document that provides the evidence, its status (exists, partial, missing) and its owner.
Deliverable: prioritized gap table, with an owner and a due date per line.
3. Build the threat model and risk assessment
Start from the architecture views: trust boundaries, data flows, assets to protect (therapeutic function, patient data, keys, firmware). Apply a recognized method (STRIDE, attack trees) and rate each scenario on exploitability: required access, skills, tooling, detectability. FDA-recognized standards such as AAMI SW96 and AAMI TIR57 provide the framework. Then link each scenario to any patient impact in the ISO 14971 analysis.
Deliverable: threat model, cybersecurity risk assessment and the four architecture views.
4. Produce the SBOM and assess known vulnerabilities
Generate the SBOM as close as possible to the shipped binary, ideally from the build pipeline, in a machine-readable format (SPDX or CycloneDX). Do not forget the layers that tools miss: embedded operating system, bootloader, Bluetooth stack, radio module firmware, statically linked libraries. Then match each component against vulnerability databases and document, for each vulnerability, whether it applies, why, and how it is handled.
Deliverable: machine-readable SBOM and known-vulnerability assessment table.
5. Have the device tested by an independent team
The FDA expects testing by people independent of the development team, described in the report (independence, expertise). The penetration test scope should follow from the threat model: every major scenario should be attempted. On a connected device, this typically covers hardware (debug ports, memory extraction), firmware (secure boot, updates), radio links (BLE, Wi-Fi, NFC), the mobile app and the cloud back end.
Schedule the test early enough to fix findings, then have the fixes verified. A report showing vulnerabilities found, fixed and retested is more reassuring than a blank one.
Deliverable: test report, remediation plan, retest report.
6. Trace, assemble and prepare postmarket
Build the traceability matrix threats → controls → tests → results, then write the postmarket cybersecurity management plan and the cybersecurity part of the labeling. Finally, have the whole file read by someone who was not on the project, asked to answer one question: "if all I had were these documents, would I be convinced?"
Deliverable: traceability matrix, postmarket plan, labeling, complete eSTAR file.
Templates for your documents
The following tables are deliberately simple. Copy them into your own document templates and adapt the columns to your quality system.
Traceability matrix
| Threat | Risk (exploitability) | Control | Requirement | Test | Result |
|---|---|---|---|---|---|
| TH-03 Malicious firmware installed through the update process | High | Firmware signature verified by the bootloader | SR-12 | PT-07, VT-21 | Pass (report v2) |
| TH-08 Dosing command replayed over BLE | Medium | Authenticated pairing, anti-replay counter | SR-04, SR-05 | PT-11 | Fixed and retested |
| TH-14 Shell access through the UART port | Low (physical access) | Port disabled in production | SR-19 | PT-02 | Pass |
Minimum SBOM fields
| Field | Example |
|---|---|
| Component and version | mbedTLS 3.6.2 |
| Supplier | Open source project / vendor |
| Unique identifier | purl or CPE |
| Relationship | Included in main firmware |
| Level of support | Actively maintained / no longer maintained / unsupported |
| End-of-support date | Known date or "not disclosed", with the related strategy |
| Known vulnerabilities | Reference to the assessment table (applicable or not, rationale) |
What the penetration test report should contain
- identity of the test team, evidence of its independence from development and of its expertise;
- exact version of the device, firmware, app and back end tested;
- scope, duration, approach (black, grey or white box) and methods used;
- mapping between tested scenarios and the threat model;
- each finding with its description, evidence, risk level and recommendation;
- status of each finding after remediation and retest results;
- rationale for unfixed findings, linked to the residual risk assessment.
Outline of the postmarket management plan
- monitoring sources (vulnerability databases, component supplier advisories, ISAOs such as Health-ISAC);
- coordinated disclosure process: public page, contact address, acknowledgement timeline;
- vulnerability assessment and prioritization method, linked to the risk assessment;
- target remediation timelines, including out-of-cycle patches for critical vulnerabilities;
- update deployment and customer communication arrangements;
- tracked metrics (time from identification to patch, deployment time, share of vulnerabilities remediated);
- end-of-support management for components and for the device.
Twelve field-tested tips
- Fill in the eSTAR at the start, not the end. It doubles as a work plan: each field points to evidence you need to produce.
- Use a Pre-Submission (Q-Sub). For an unusual architecture or a structuring choice, getting the FDA's view upfront avoids a bad surprise during review. Allow two to three months.
- Write for a busy reader. Open each document with a one-page summary pointing to detailed sections and matrix identifiers.
- Number everything. Threats, requirements, tests, findings: stable identifiers make traceability verifiable and answers to questions fast.
- Look beyond the enclosure. The service tool, the factory programming bench and the cloud admin account are attack surfaces the FDA expects to see addressed.
- Think "multi-patient". Show what prevents a cloud or update compromise from hitting the entire fleet at once.
- Justify unfixed vulnerabilities. "Not applicable" must be demonstrated: feature not compiled, interface absent, exploitation condition impossible.
- Freeze the tested version. The test report must cover the submitted version, or explain precisely the differences and why they have no security impact.
- Ask for a usable report. Before signing with a test provider, ask for a sample report and check it covers the items listed above.
- Leave time to fix. A test run two weeks before submission leaves two options: submit with open vulnerabilities or delay submission.
- Make updates credible. Describe how a patch reaches a device installed at a patient's home or in a hospital, how long it takes, and what happens if it fails.
- Pool with Europe. The same evidence feeds the MDR file (MDCG 2019-16) and, for your non-medical products, the Cyber Resilience Act regulation. Structure it once for both markets.
After submission: answering FDA questions
Even a good file often receives questions, in the form of an Additional Information request. Review is then put on hold and the response time is limited (180 days for a 510(k)): past that, the submission is considered withdrawn.
- Answer the question asked, and only that question. Restate the request, provide the evidence, say where it sits in the file.
- Prefer evidence over argument. A configuration extract, a test result or a matrix line convinces more than a paragraph of explanation.
- If additional testing is needed, start it immediately. It is the longest task; a provider already familiar with the device will save weeks.
- Capitalize. Each question received is a gap in your document templates: fix the template for the next submission.
On the inspection side, remember that cybersecurity records are part of the quality system: an investigator may ask for the version history of the threat model, how a reported vulnerability was handled, or the corrective actions that followed. What was promised in the submission must be applied, and evidenced, after market entry.
Pre-submission checklist
- "Cyber device" status confirmed and justified
- Cybersecurity owner for the file appointed
- Global system diagram and interface inventory up to date
- eSTAR cybersecurity section filled in and reviewed
- Threat model covering device, apps, cloud, service and manufacturing
- Exploitability-based risk assessment, linked to ISO 14971
- Four architecture views: global, multi-patient, updatability, use cases
- Machine-readable SBOM with level and end of support for each component
- Known vulnerabilities assessed one by one
- Unresolved software anomalies assessed for security impact
- Independent penetration test on the submitted version, fixes retested
- Complete traceability matrix threats → controls → tests → results
- Cybersecurity labeling written
- Postmarket plan with CVD, remediation timelines and metrics
- Full review by someone outside the project