---
title: "AI/ML Medical Device Cyber Walkthrough - PCCP, SPDF, MDR &amp;…"
description: "Interactive walkthrough for AI/ML medical devices: identify cyber risks and pick the right FDA SPDF/PCCP, EU MDR (MDCG 2019-16 Rev.2), and EU CRA sections for…"
lang: en
json-ld: |
  [
    {
      "@context": "https://schema.org",
      "@graph": [
        {
          "@type": "WebSite",
          "@id": "https://medtechcybertips.com/#website",
          "url": "https://medtechcybertips.com/",
          "name": "MedTech Cyber Tips",
          "description": "The ultimate organized guide to medical device cybersecurity.",
          "inLanguage": "en-US",
          "publisher": {
            "@id": "https://medtechcybertips.com/#org"
          },
          "potentialAction": {
            "@type": "SearchAction",
            "target": "https://medtechcybertips.com/topics?q={search_term_string}",
            "query-input": "required name=search_term_string"
          }
        },
        {
          "@type": "Organization",
          "@id": "https://medtechcybertips.com/#org",
          "name": "MedTech Cyber Tips",
          "url": "https://medtechcybertips.com/",
          "logo": "https://medtechcybertips.com/icon-512.png",
          "sponsor": {
            "@type": "Organization",
            "name": "Blue Goat Cyber",
            "url": "https://bluegoatcyber.com",
            "sameAs": [
              "https://home.bluegoatcyber.com/",
              "https://codebluechart.com/",
              "https://why.bluegoatcyber.com/",
              "https://spdf.bluegoatcyber.com/",
              "https://threatmodel.bluegoatcyber.com/",
              "https://pentesting.bluegoatcyber.com/",
              "https://premarket.bluegoatcyber.com/",
              "https://fdaresponse.bluegoatcyber.com/",
              "https://postmarket.bluegoatcyber.com/",
              "https://goatwatch.bluegoatcyber.com/"
            ]
          }
        }
      ]
    },
    {
      "@context": "https://schema.org",
      "@type": "HowTo",
      "name": "AI/ML medical device cybersecurity walkthrough",
      "description": "Guided walkthrough that helps MedTech teams identify AI/ML cyber risks and select the right PCCP/SPDF, EU MDR, and EU CRA artifacts for their submission and postmarket program.",
      "datePublished": "2026-07-10",
      "dateModified": "2026-07-10",
      "url": "https://medtechcybertips.com/aiml-walkthrough",
      "step": [
        {
          "@type": "HowToStep",
          "name": "Identify device modality"
        },
        {
          "@type": "HowToStep",
          "name": "Classify learning behavior (locked, PCCP, continuous)"
        },
        {
          "@type": "HowToStep",
          "name": "Map data flow (on-device, cloud, federated)"
        },
        {
          "@type": "HowToStep",
          "name": "Select target jurisdictions (FDA, EU MDR, EU CRA)"
        },
        {
          "@type": "HowToStep",
          "name": "Prioritize AI/ML cyber risks"
        },
        {
          "@type": "HowToStep",
          "name": "Generate tailored SPDF / PCCP / MDR / CRA artifact list"
        }
      ]
    }
  ]
---

[Skip to main content](#main)

[

MedTech Cyber Tips

The Ultimate Guide



](/)

[Home](/)[Find Your Path](/journey)[All Topics](/topics)Reference[About](/about)

Search ⌘K

Guided walkthrough · 2026 edition

# AI/ML medical device cyber walkthrough

Answer five short questions and get a tailored list of AI/ML cyber risks and the exact PCCP / SPDF sections (FDA Feb 3, 2026 guidance), MDR / MDCG mappings, and EU CRA obligations to include in your technical file.

Last updated Jul 10, 2026· 1mo ago  ~2 minutes · no login 

Step 1 of 5

## What's the device modality?

Modality drives the highest-signal adversarial threats and the evaluation evidence you'll need.

Medical imaging (CT/MR/US/X-ray)

CNN/vision transformer on DICOM

Signals / physiological (ECG, EEG, PPG)

Time-series classifier or anomaly detector

Clinical NLP / LLM assistant

Text summarization, coding, or clinical decision support

Genomics / omics

Variant calling, risk scoring

Wearable / at-home monitor

On-device inference, intermittent connectivity

Robotics / surgical / closed-loop

Real-time control with safety envelope

Step 2 of 5

## How does the model change after clearance?

This determines whether you need a PCCP and how tight the change-control envelope has to be.

Locked model (no post-market changes)

Frozen weights; standard 510(k)/PMA path

PCCP-managed adaptive

Pre-specified modifications & protocol under Section 524B / Feb 3 2026 guidance

Continuously / online learning in field

Highest scrutiny - requires very tight change controls

Step 3 of 5

## Where does inference happen?

Data-flow topology decides which cloud, transport, and cryptographic controls you must document.

On-device inference only

Model artifact ships with device

Cloud inference / SaMD API

Data leaves device to model service

Federated / split learning

Weights or gradients travel; data stays local

Step 4 of 5

## Which jurisdictions do you ship into?

Pick every regime you'll be reviewed under - obligations stack and CRA has separate reporting.

United States (FDA)

Section 524B FD&C + Feb 3 2026 guidance

European Union (MDR)

MDR Annex I §17.2 + MDCG 2019-16 Rev.2

EU CRA-in-scope accessories / companion apps

Regulation (EU) 2024/2847 - reporting from Sept 11 2026

Select all that apply.

Step 5 of 5

## Which risks are you already worried about?

Optional - we always compute a baseline. Selecting risks here raises their priority in the output.

Adversarial inputs

Crafted perturbations (imperceptible noise, prompt jailbreaks, out-of-distribution triggers) cause misclassification or unsafe action.

Training-data / supply-chain poisoning

Malicious contribution to training, fine-tuning, or upstream dataset (labels, backdoors, dependency confusion via ML libraries).

Model / weight extraction

Query-based or physical readback of proprietary weights, feature representations, or training data.

Silent performance drift

Population shift, sensor firmware change, or PCCP modification degrades sensitivity/specificity without triggering an alarm.

PHI / PII leakage

Membership inference, log leakage, or over-collection through inference pipeline exposes protected health information.

Prompt injection / tool abuse (LLM only)

Untrusted clinical text overrides system prompt, extracts tools, or exfiltrates PHI to attacker-controlled endpoints.

Select all that apply.

Complete steps 1–4 to see recommendations.  Reset

MedTech Cyber Tips

The organized, end-to-end guide to medical device cybersecurity, from concept through postmarket. Part of the Blue Goat Cyber family.

Topics

-   [Overview](/topics/home)
-   [Why It Matters](/topics/why)
-   [SPDF](/topics/spdf)
-   [Threat Modeling](/topics/threatmodel)
-   [Pentesting](/topics/pentesting)

Reference

-   [Glossary](/glossary)
-   [FDA vs MDR](/compare)
-   [Resources](/resources)
-   [Updates](/updates)
-   [RSS feed](/rss.xml)

© 2026 medtechcybertips.com. An educational resource sponsored by Blue Goat Cyber.

Not legal or regulatory advice.