Safeguard
Supply Chain Security•

How Medical Devices and Wearables Get Hacked

How medical devices get hacked: insulin pumps, pacemakers, infusion pumps, hospital IoT, fitness trackers. Real cases and how to secure them.

Safeguard Security Team
68 min read

Medical devices and wearables are rarely hacked the way films show it. The documented risks are radio links that never checked who was talking, hospital equipment running operating systems that stopped getting updates, network libraries shared by hundreds of products, and phone apps and cloud services that leak the health data the device collects. Every one of those has a known fix, and this guide walks through them calmly, one at a time.

Insulin pumps, pacemakers, infusion pumps, patient monitors, fitness trackers and smartwatches are all small computers with a radio. Some keep a person alive. Others record their heart rate, their sleep and the route they run every morning. A vulnerability in a pump and a leak from a running app are very different problems, so this guide keeps them apart and is honest about how likely each one is.

It is written for three groups of readers. Patients and everyday users who want to know what is actually risky and what to do about it. Hospital IT, clinical engineering and security teams who look after connected equipment. And the engineers and product security teams at device makers, who will find the regulation, SBOM and supply chain sections most useful. Jargon is defined on first use.

This is a defender's guide written from the attacker's point of view. For each attack class you will find how it works and why, what an attacker needs, documented cases with CVE numbers where they exist, the impact, how to detect it and how to defend against it. There are no exploit steps, payloads or tool command lines. A note on the law: testing a device you do not own, or do not have written permission to test, is illegal in most countries. With medical devices it can also put a patient at risk. Tools and research teams are named here only as context for authorised work.

This post is part of a device security series. The companion guides cover how Bluetooth gets hacked, how phones get hacked, how smart home devices get hacked and how Wi-Fi gets hacked. For where device hardware and firmware come from, see how a laptop is built and fabless chip designers and the software supply chain.

Contents

  1. How do medical devices and wearables get hacked? The short answer
  2. How do medical devices and wearables work, and what software do they run?
  3. What is the attack surface of a medical device or wearable?
  4. What are the main ways medical devices and wearables get hacked?
  5. Attack 1: Can insulin pumps be hacked over the air?
  6. Attack 2: Can pacemakers and their programmers be hacked?
  7. Attack 3: How are infusion pumps attacked?
  8. Attack 4: How does one network library put thousands of devices at risk?
  9. Attack 5: Why is hospital IoT running old software such an easy target?
  10. Attack 6: How is Bluetooth LE in wearables attacked?
  11. Attack 7: How do companion apps and cloud services leak health data?
  12. Attack 8: How do wearables expose privacy when nobody hacks them?
  13. Why can't hospitals just patch medical devices?
  14. How can you tell if a medical device or wearable has been compromised?
  15. What do HIPAA and GDPR mean for health data from devices?
  16. What do the EU MDR, IEC 81001-5-1 and the CRA expect from device makers?
  17. Hardening checklists for patients, hospitals and makers
  18. Where does the medical device software supply chain break?
  19. How Safeguard helps medical device and wearable makers
  20. The takeaway
  21. Frequently asked questions
  22. Sources

How do medical devices and wearables get hacked? The short answer

Medical devices and wearables get hacked in eight main ways: unauthenticated radio links on pumps and implants, weak update and programming paths, flaws in hospital device software, shared network libraries such as Treck and IPnet, legacy operating systems in hospitals, Bluetooth LE weaknesses in wearables, leaky companion apps and clouds, and privacy exposure built into the product itself.

Two things are worth saying up front, because they set the tone for everything that follows.

First, the publicly documented attacks on implanted and body-worn devices are research demonstrations, not crimes. In the cases covered here, the makers that commented reported no confirmed exploitation. Medtronic's own letter on its older insulin pumps says it had "received no confirmed reports of unauthorized people changing settings or controlling insulin delivery." That matters. A patient should never stop using a device that keeps them well because of a research paper, and every maker cited here says the same.

Second, the attacks that have caused real disruption are the boring ones. Ransomware on unpatched hospital computers. Health data left in an open database. A cloud service going offline. They hit thousands of people at once and need far less skill than a radio attack on a single pump.

So the defences sit in four places:

  • The device and its radio need authentication, encryption and signed updates, built in by the maker.
  • The hospital network decides whether a compromised PC can reach a pump or a scanner. Segmentation and monitoring do most of the work.
  • The app and the cloud hold most of the data. Unique passwords, two-factor authentication and careful sharing settings protect it.
  • The supply chain decides how fast a fix arrives. Makers who know every component in every build can ship one in weeks rather than years.

How do medical devices and wearables work, and what software do they run?

A medical device or wearable is a small computer built around a microcontroller or system on chip, running a real-time operating system or embedded Linux, with a radio stack from the chip vendor and the maker's own application on top. Hospital devices often run embedded RTOSes or Windows. A phone app and a cloud service usually sit around them.

Here is what that means for each device type in this guide.

  • Insulin pumps deliver insulin through a small motor, following a basal schedule and doses (boluses) the user requests. Most talk wirelessly to a continuous glucose monitor (CGM), a blood glucose meter, a remote or a phone app. Older pumps use proprietary sub-gigahertz radios; newer ones use Bluetooth Low Energy (BLE).
  • Pacemakers and implantable cardioverter defibrillators (ICDs) sit under the skin and regulate or restart the heartbeat. They talk by short-range telemetry to a clinician programmer in the clinic and to a bedside home monitor, which uploads readings to the clinic over the phone network or internet.
  • Infusion pumps in hospitals deliver drugs and fluids at set rates. Smart pumps carry a drug library, a list of approved drugs and dose limits, which is loaded from a central server over the hospital network.
  • Hospital IoT covers everything else that is connected: patient monitors and their central stations, imaging systems such as MRI and CT scanners with their Windows workstations, lab analysers, pneumatic tube systems that move samples, nurse call and building systems.
  • Fitness trackers and smartwatches measure steps, heart rate, sleep and location. They sync over BLE to a phone app, which uploads the data to the maker's cloud and often shares it with other apps.

Three-column diagram of the software stacks in body-worn or implanted devices, hospital devices and the surrounding app, cloud, programmer and hospital server systems

Figure 1. Three typical stacks. In each, most layers come from suppliers rather than the brand on the label.

Reading the stacks from the bottom up:

  1. Hardware. A low-power microcontroller (MCU) or system on chip (SoC) with radio, flash memory and sometimes a secure element that stores keys. Hospital devices may use a full PC board.
  2. Operating system. On small devices, a real-time operating system (RTOS), a minimal OS designed for predictable timing. On hospital equipment, embedded RTOSes such as VxWorks, OSE or ThreadX, or a general-purpose Windows or Linux build.
  3. Supplier libraries and SDKs. The chip vendor's software development kit (SDK) with its BLE stack, a TCP/IP network stack, cryptography, logging, databases and viewers. The device maker usually licenses these, often as compiled binaries.
  4. The maker's application. Dosing logic, alarms, therapy settings, image acquisition, activity tracking.
  5. The surrounding systems. Companion apps, cloud APIs, clinician programmers, home monitors, drug library servers and links into the electronic health record (EHR).

Every layer has its own supplier and its own update cycle. That single fact explains most of this guide: why one network library flaw reached infusion pumps and patient monitors from different makers, and why a fix can take years to reach a patient.

What is the attack surface of a medical device or wearable?

The attack surface of a medical device or wearable is every way to reach its code or data: the radio link and pairing, the clinician programmer and home monitor, physical ports, the hospital network, the update channel, the companion app, the maker's cloud account and the third-party components inside the firmware.

Map of ten entry points around a medical device, coloured by whether the attacker must be nearby, on the network, on the internet or inside the supply chain

Figure 2. Ten entry points. The colour shows what the attacker needs to reach each one.

The surface splits by distance, and distance is what separates a realistic risk from a laboratory demonstration.

SurfaceWho can reach itTypical weakness
Radio link and pairingSomeone within radio rangeNo authentication, no encryption, no replay protection
Clinician programmer and home monitorSomeone with physical access, or the network path for updatesUnsigned updates, unencrypted patient data on disk
Physical portsSomeone holding the deviceConfiguration changes without authentication
Hospital networkAnyone on the clinical network, or malware on a connected PCLegacy operating systems, flat networks, unauthenticated protocols
Update channelWhoever controls the network path or update serverUpdates without signatures or over plain HTTP
Companion appAnyone who can get code or a malicious link onto the phoneWeak pairing, stored tokens, over-broad data sharing
Maker cloud and accountAnyone on the internetReused passwords, exposed databases, outages
Supplier componentsThe component's own attackersShared TCP/IP and BLE stacks, old libraries

Bar chart showing how close an attacker must be for implant telemetry, pump and wearable radios, hospital networks, apps and cloud, and shared components

Figure 3. The further away an attacker can be, the more people can try. Implant telemetry needs someone within a few metres; a shared component flaw reaches every product that ships it.

What are the main ways medical devices and wearables get hacked?

The eight attack classes below cover nearly every publicly documented medical device and wearable security case. Radio attacks on a single pump or implant need skill and proximity and remain research demonstrations. Ransomware on legacy hospital systems, exposed cloud data and shared library flaws need less skill and affect far more people.

Qualitative matrix plotting attack classes by attacker effort and number of people or devices affected

Figure 4. A qualitative view of effort against scale, drawn from the cases in this guide. It is an editorial judgement, not a measured dataset.

The same patterns have repeated for a decade, and regulation has followed.

Timeline of medical device and wearable security research and incidents from 2016 to 2027

Figure 5. Ten years of medical device and wearable security events, ending with the EU Cyber Resilience Act deadlines.

Each attack section uses the same structure: how it works, what the attacker needs, real cases, impact, detection and defence.

Attack 1: Can insulin pumps be hacked over the air?

Yes, some older insulin pumps can be controlled over the air by someone nearby, because their radio protocol never checked who was sending commands. Researchers showed this on the Animas OneTouch Ping in 2016 and Medtronic MiniMed 508 and Paradigm pumps in 2019. No confirmed attacks on patients have been reported. Newer pumps authenticate and encrypt.

How it works and why

Many insulin pumps accept commands from a remote, a meter or a phone. On older designs, that wireless conversation had three gaps. It was not encrypted, so anyone listening could read it. It did not properly authenticate the sender, so the pump could not tell a real remote from an impostor. And it had no replay protection, meaning no sequence numbers or timestamps, so a recorded command could be played back later and the pump would accept it again.

Why did designs like this ship? Many were designed before connected medical devices were seen as a security target, under tight battery and size limits, and with the assumption that short radio range was protection enough.

What the attacker needs

Radio equipment for the pump's frequency, technical knowledge of the protocol, the pump's serial number in some designs, and physical proximity to the patient. For the MiniMed 600 series issue described below, the attacker also had to be nearby at the moment the pump was being paired.

Real cases

  • Animas OneTouch Ping (2016). Jay Radcliffe of Rapid7 published advisory R7-2016-07. The meter, which doubles as a remote, talked to the pump over 900 MHz radio in cleartext (CVE-2016-5084), used a weak pairing process whose key was itself sent in the clear (CVE-2016-5085), and had no replay protection, so captured commands could be replayed to trigger a dose (CVE-2016-5086). Normal range was about 10 metres, though the researchers believed specialist equipment could extend it much further. Rapid7 said the risk of wide-scale exploitation was "relatively low" and that the disclosure was not "cause for panic." The fix for worried users was simple: turn off the pump's radio feature and follow the maker's guidance, including the bolus limits that can be set on the pump itself.
  • Medtronic MiniMed 508 and Paradigm series (2019). On 27 June 2019 Medtronic issued an urgent field safety notification. Its January 2023 follow-up letter explains that "an unauthorized person with special technical skills and equipment could potentially connect wirelessly to a nearby insulin pump to change settings and control insulin delivery." The CVE record, CVE-2019-10964, says the radio protocol "does not properly implement authentication or authorization." These pumps could not be fixed with a software update, so Medtronic moved patients to newer models and published ten precautions for anyone still using one. A separate recall of the optional MiniMed remote controllers, first communicated in August 2018, was expanded in October 2021, according to the same letter and SecurityWeek.
  • Medtronic MiniMed 600 series (2022). Medtronic's own internal testing found that, under specific circumstances, a nearby person could interfere with the wireless pairing between pump system components. The security bulletin says "this cannot be done over the internet," tracks it as CVE-2022-32537 with a CVSS score of 4.8, and recommends turning off the Remote Bolus feature, which was on by default, and pairing devices in private. Medtronic said it had no evidence the issue had occurred.

Impact

In the worst case, a delivered dose the patient did not ask for (risking low blood sugar, hypoglycaemia) or stopped delivery (risking high blood sugar and diabetic ketoacidosis). Medtronic's letter names both. The data exposure from eavesdropping, mainly glucose and dose readings, is less serious but still personal health data.

Detection

  • Pump alerts and alarms are the main signal. Unexpected boluses, changed settings or connection requests you did not start are all reasons to act.
  • Look at the pump's history screen regularly for doses you do not recognise.
  • On phone-connected pumps, the app's list of paired devices and remote features should match what you set up.

Defence

For patients:

  • Keep using your pump. The risk from stopping therapy is far greater than the risk of an attack.
  • Read and act on any safety notice from the maker or your clinic. Ask your diabetes team whether your model is affected.
  • Turn off remote features you do not use, such as Remote Bolus or a radio remote.
  • Keep the pump and connected parts within your control, pair devices at home rather than in public, and do not share serial numbers beyond your clinician and the maker.
  • Cancel any dose you did not start, check your glucose, and report it to the maker.
  • Only use apps and software the maker has authorised for your pump.

For makers: authenticate both ends of every radio link with keys unique to each device, encrypt the traffic, add replay protection with counters or timestamps, protect pairing with an out-of-band step, and make remote dosing opt-in. Design in a secure update path, because a pump that cannot be updated can only be replaced.

Attack 2: Can pacemakers and their programmers be hacked?

Research has shown that pacemakers, ICDs and especially the programmers and home monitors around them had serious weaknesses, including unauthenticated telemetry and unsigned software updates. Makers have shipped firmware updates and switched off risky update paths. No cyberattack on a patient's implant has been reported, and attacks need close proximity and specialist knowledge.

How it works and why

An implant is the hardest part of the system to reach. It talks only over short-range telemetry, and on many designs only when a clinician activates it. The systems around it are easier targets:

  • Clinician programmers are portable computers that read and change implant settings. They run full operating systems with many third-party libraries and receive software updates.
  • Home monitors sit by the bed, talk to the implant overnight, and upload readings to the clinic.
  • Telemetry protocols decide whether the implant checks who is talking to it.

If a programmer accepts an unsigned update, an attacker who can tamper with that update path can change the programmer's software. If the telemetry protocol has no authentication, a nearby device that speaks the protocol can read or change settings while telemetry is active.

What the attacker needs

For telemetry attacks: proximity, typically within a few metres; a radio that speaks the proprietary protocol; deep knowledge of the device and of cardiac electrophysiology; and timing, because telemetry is only active at certain moments. For programmer update attacks: a position on the network path between the programmer and the update server.

Real cases

  • WhiteScope ecosystem study (2017). Billy Rios and Jonathan Butts of WhiteScope examined pacemakers, ICDs, home monitors and programmers from four makers. SecurityWeek reported that the four programmers contained more than 300 third-party libraries, 174 of which had more than 8,000 known vulnerabilities between them. All the programmers examined had outdated software, some stored unencrypted patient data, and home monitors used hardcoded credentials and firmware updates that were not digitally signed. The results of that study are the clearest public evidence that implant ecosystems are, first of all, a software supply chain problem.
  • Abbott pacemakers (2017). Abbott, which had acquired St. Jude Medical, released a firmware update for about 465,000 pacemakers across six model families, according to CSO Online. The update addressed an authentication weakness that could let a nearby attacker send commands over radio, tracked as CVE-2017-12712, along with battery depletion and unencrypted data issues. It took about three minutes and was applied in clinic, and Abbott noted a very low risk of the update itself malfunctioning. That trade-off, updating an implant in a person's chest, is the patching problem in its sharpest form.
  • Medtronic CareLink 2090 programmer (2018). The programmer downloaded software updates over a VPN but requested them over plain HTTP and did not check that the VPN was still connected first (CVE-2018-10596). Rios and Butts presented the research at Black Hat in August 2018. In October 2018 Medtronic switched off internet-based updates for about 34,000 programmers, moving them to USB updates from the clinic, and said it had received no reports of exploitation (TechCrunch). A related record, CVE-2018-18984, covers patient data that was not sufficiently encrypted at rest on CareLink and Encore programmers.
  • Medtronic Conexus telemetry (2019). Conexus, the radio protocol in some Medtronic ICDs and cardiac resynchronisation defibrillators and their monitors and programmers, did not implement authentication (CVE-2019-6538) or encryption (CVE-2019-6540). Medtronic's bulletin explains the real-world limits: telemetry is activated by a clinician in the same room, activation outside the clinic is limited and hard to predict, range typically does not exceed 6 metres, and the issue is not reachable from the internet. Medtronic pacemakers do not use Conexus. Updates were applied automatically at clinic visits between January 2020 and June 2021. Medtronic reported no cyberattack, privacy breach or patient harm.

Impact

In theory, changed therapy settings or a drained battery. In practice, the documented impact has been exposure of patient data stored on programmers, and the cost and anxiety of large update campaigns. The stakes are high enough that makers treat these as safety issues, even with no attacks recorded.

Detection

  • For patients: unexplained device behaviour, alerts from the home monitor, or a clinic telling you readings look unusual. Ask at each check-up whether your device has the latest firmware.
  • For clinics: keep programmers on supported software, track maker bulletins, and record which firmware each patient's implant runs.
  • For makers: monitor fleet telemetry for unusual programming sessions, and keep logs on programmers and home monitors.

Defence

For patients: keep your follow-up appointments, because that is where updates are applied. Use only the home monitor your clinic or the maker provided, keep it plugged in and in your control, and report anything unusual. Medtronic gives the same advice.

For clinics and hospitals: treat programmers as managed computers. Keep them updated through the maker's approved route, encrypt or restrict their storage, never connect them to general networks, and wipe them before disposal, since WhiteScope found unencrypted patient data stored on programmers.

For makers: authenticate and encrypt telemetry, limit when it can be activated, sign every update and verify signatures before installing, keep an inventory of every library in the programmer and home monitor, and patch those libraries on a schedule. Thousands of known vulnerabilities in third-party code is exactly what a software bill of materials (SBOM) and continuous component monitoring are designed to catch.

Attack 3: How are infusion pumps attacked?

Hospital infusion pumps are attacked mainly through the network: the links between pumps, drug library servers and management software. Documented vulnerabilities include unsigned firmware for wireless modules, configuration changes without authentication, unprotected data between pump modules, and outdated third-party libraries in the server software. None has been reported exploited.

How it works and why

A smart infusion pump is one part of a system. The pump talks to a server that sends it the drug library and collects infusion records, and the hospital's staff manage both through software on Windows servers. That gives an attacker several targets: the network session between pump and server, the files the server distributes, the pump's own configuration interface, and the server software itself.

Pumps are attractive to researchers because a wrong infusion rate is a direct safety risk, and because pumps are numerous, mobile and often on the hospital's Wi-Fi.

What the attacker needs

Access to the clinical network for network-based issues; physical access to the pump for configuration interface issues; or a foothold on the management server.

Real cases

  • BD Alaris PC Unit and Systems Manager (2020). A network session authentication vulnerability between the BD Alaris PC Unit model 8015 (version 9.33.1 and earlier) and BD Alaris Systems Manager (4.33 and earlier) could let an attacker modify configuration headers of data in transit and cause a denial of service on the pump, according to the CVE record CVE-2020-25165.
  • BD Alaris System with Guardrails Suite MX (2023). BD's bulletin lists eight vulnerabilities that BD found itself through routine internal testing. They include a wireless card firmware update package that "is not properly signed and can be modified" (CVE-2023-30559), pump configuration that can be changed without authentication through a physical connection (CVE-2023-30560), unprotected data between the pump and its modules, dataset files that could be tampered with, and two web flaws in Systems Manager. One more is telling: an outdated Apache log4net library in BD's Calculation Services (CVE-2018-1285), a 2018 flaw in a third-party logging component, still present in a 2023 medical product. BD released system version 12.3, which fixed several of the issues, and states there have been no reports of exploitation. Scores were mostly medium.
  • URGENT/11 in an infusion pump (2019). A hospital using Armis monitoring found that a BD Alaris pump ran the OSE operating system with the IPnet network stack, making it affected by the URGENT/11 vulnerabilities covered in the next section (Armis via PR Newswire).

BD deserves credit here. Discovering eight vulnerabilities in your own product, assigning CVEs as a CVE Numbering Authority and publishing them is what mature product security looks like. Calm, published, fixed.

Impact

Denial of service on a pump, tampered drug libraries that could change dose limits, or exposure of infusion records. Pumps keep running their current programme, and clinical staff check infusions at the bedside, which limits the safety impact of many network-level issues.

Detection

  • Monitor pump network segments for connections from anything other than the drug library server and management systems.
  • Alert on drug library changes outside the change process, and verify library versions on pumps against the approved version.
  • Check pump firmware and server versions against maker bulletins every quarter.

Defence

  • Segment pumps onto their own network zone that can reach only the pump server. Figure 7 later in this guide shows the layout.
  • Apply maker updates for both pumps and server software on a planned schedule, following the maker's instructions.
  • Restrict physical access to pump service ports and keep pumps in clinical areas.
  • Control drug library changes with approval, logging and version checks.
  • For makers: sign every firmware component including wireless modules, authenticate configuration interfaces, protect module-to-module data, and track the versions of every third-party library in server software. An SBOM would have flagged a 2018 log4net vulnerability years earlier.

Attack 4: How does one network library put thousands of devices at risk?

Many medical devices use the same few embedded TCP/IP network stacks, licensed from suppliers. When researchers found Ripple20 in the Treck stack in 2020 and URGENT/11 in the IPnet stack in 2019, the flaws reached infusion pumps, patient monitors and other devices from many makers at once, and most makers first had to work out whether they were affected.

How it works and why

Every networked device needs a TCP/IP stack, the code that handles internet protocol packets. Writing one is hard, so device makers license an existing stack, often as part of an RTOS or a chip vendor's SDK. The same stack then ships in printers, industrial controllers, infusion pumps and patient monitors.

Packet parsing code is a classic source of memory corruption bugs. A malformed packet that the stack mishandles can crash the device or, in the worst case, let an attacker run code. Because the stack sits at the lowest level, the flaw is reachable by anyone who can send packets to the device.

Diagram showing three shared components, Treck TCP/IP, IPnet and chip vendor BLE SDKs, flowing through operating systems, SDKs and modules into many types of medical and consumer devices

Figure 6. How one shared component reaches products from dozens of makers.

What the attacker needs

Network access to the device, usually from the same network segment, and for remote code execution, deep knowledge of the specific stack and device.

Real cases

  • Ripple20 (June 2020). Researchers at JSOF disclosed 19 vulnerabilities in the Treck TCP/IP library, which HIPAA Journal reported was present in hundreds of millions of devices across nearly every sector. Two were rated 10 out of 10, including CVE-2020-11896, remote code execution through malformed IPv4 tunnelling packets. Medical infusion pumps were among the affected device types, and Baxter and B. Braun were named among affected companies. Treck's fix was version 6.0.1.67 or later. The name describes the problem: a flaw in one small component rippling out through many companies' products. Our post on CVE-2020-11899 in the Treck stack has more on one of the 19.
  • URGENT/11 (July 2019 onward). Armis found 11 vulnerabilities in IPnet, the TCP/IP stack in Wind River's VxWorks, including CVE-2019-12256, a stack overflow when parsing IPv4 options. In October 2019 Armis reported that IPnet had also been licensed into other RTOSes, including ENEA's OSE, Green Hills INTEGRITY, Microsoft ThreadX, Mentor's Nucleus, ITRON and ZebOS (PR Newswire). That is how a BD Alaris infusion pump on OSE turned out to be affected, and why a Spacelabs Xprezzon bedside monitor was demonstrated at Black Hat. More than 30 vendors issued advisories, including GE Healthcare, Philips and Drager. Armis noted that medical devices have long development cycles and life cycles, which leaves them prone to vulnerabilities in legacy code.

Impact

Remote crash or takeover of devices on the same network, from many makers at once. The longer-lasting impact is operational: hospitals spent months asking makers whether each product was affected, and makers spent months asking their suppliers.

Detection

  • Passive network monitoring tools can fingerprint which stack a device runs and flag malformed packets aimed at it.
  • Makers can search their firmware builds for the affected component, if they have an SBOM. Without one, they have to ask suppliers or analyse binaries.
  • Hospitals should track maker advisories that reference these named vulnerability sets.

Defence

  • Hospitals: segment devices, block traffic types the device does not need (such as IP tunnelling or unusual IP options at the firewall where possible), and apply maker updates when they arrive.
  • Makers: keep a component inventory for every firmware build, including the licensed RTOS and its network stack version. Ask RTOS and SDK suppliers for their own SBOMs. When a stack vulnerability lands, the time to answer "are we affected and which builds?" should be hours, not weeks.

Attack 5: Why is hospital IoT running old software such an easy target?

Hospitals run many devices on operating systems that no longer receive security updates, because the device was validated on that version and the maker has not certified a newer one. Unit 42 found 83 percent of medical imaging devices on unsupported systems in its 2020 report. Malware that reaches a flat hospital network can spread to them.

How it works and why

An MRI scanner, a lab analyser or an imaging workstation can be in service for many years. The Windows version it shipped with may reach end of support long before the scanner does. Windows 7, for example, stopped receiving security updates from Microsoft on 14 January 2020. The hospital cannot simply upgrade the operating system, because the device was tested and approved as a whole, and an unvalidated change could affect it in unpredictable ways. The next section explains that constraint in detail.

Old systems are dangerous mainly because of what is next to them. If the network is flat, ransomware that arrives through a phishing email on a staff PC can reach an imaging workstation with a known, unpatched vulnerability.

What the attacker needs

A foothold anywhere on the hospital network, usually through a staff PC, and a network path to the device. Self-spreading malware needs nothing else.

Real cases

  • WannaCry (May 2017). The WannaCry ransomware worm spread through unpatched Windows systems worldwide. A peer-reviewed analysis by Ghafur and colleagues in npj Digital Medicine recorded 34 NHS hospital trusts in England infected, which were "locked out of their digital systems and medical devices, such as MRI scanners," and estimated £5.9 million of lost hospital activity at infected trusts during that week. No difference in mortality was found. Our WannaCry retrospective covers the technical side.
  • Unsupported operating systems (2020). Palo Alto Networks' Unit 42 2020 IoT Threat Report, based on 1.2 million IoT devices in enterprise and healthcare organisations in 2018 and 2019, found 83 percent of medical imaging devices running on unsupported operating systems, a 56 percent jump from 2018 driven by Windows 7 reaching end of life. It also found 72 percent of healthcare VLANs mixed IoT and IT assets, which lets malware spread from PCs to devices.
  • PwnedPiper (2021). Armis disclosed nine vulnerabilities in the Swisslog Translogic Nexus control panel that runs hospital pneumatic tube stations, the systems that move items between departments. Help Net Security reported the system is used in more than 3,000 hospitals worldwide. The flaws could be triggered by unauthenticated network packets. Firmware version 7.2.5.7 fixed all but one, an unsigned firmware upgrade path, which Swisslog said would be fixed in a later release. It is a good reminder that hospital IoT includes building systems nobody thinks of as medical devices.

Impact

Cancelled appointments and diverted care when systems are locked or switched off, as the WannaCry study shows. Ransomware incidents in hospitals are the most common way cyber risk actually reaches patients today.

Detection

  • An accurate inventory of every connected device with its operating system version is the starting point. Passive network monitoring builds one without disrupting devices.
  • Alert on internal scanning, new remote access connections and unusual protocols on device segments.
  • Watch for devices that suddenly start talking to the internet.

Defence

Hospital network diagram with corporate IT separated by an internal firewall from zones for infusion pumps, patient monitors, imaging and building systems, plus controlled vendor access and a monitoring sensor

Figure 7. A segmented hospital network. Each device class gets its own zone, and zones talk only through allow-listed paths.

  • Segment by device class. Put pumps, monitors, imaging, lab and building systems in their own zones. Allow only the flows each class needs, and block device-to-PC traffic by default.
  • Isolate unsupported systems. If a scanner must stay on an old operating system, give it the smallest possible network exposure. EU medical device guidance gives the same advice: avoid end-of-life components where possible, and otherwise add measures such as network isolation (MDCG 2019-16).
  • Control vendor remote access through a monitored jump host with named, time-limited accounts.
  • Plan replacement. Put end-of-support dates for operating systems into capital planning, and ask makers for supported upgrade paths before buying.
  • Rehearse downtime. Know how each department works if its devices or systems are offline for a week.

For more on segmentation in a home context, the smart home guide covers the same idea with consumer routers.

Attack 6: How is Bluetooth LE in wearables attacked?

Bluetooth LE in wearables is attacked through flaws in the chip vendor's BLE stack, through pairing modes that do not authenticate the other device, and through fixed radio addresses that let a tracker be followed. The SweynTooth research in 2020 crashed or deadlocked BLE chips used in fitness trackers and flagged medical products using the same chips.

How it works and why

Bluetooth Low Energy (BLE) is the low-power version of Bluetooth used by nearly every fitness tracker, smartwatch, CGM and modern insulin pump. Three properties matter for security:

  • The stack comes from the chip vendor. The BLE software that parses radio packets is part of the chip vendor's SDK. If it mishandles a malformed packet, every product built on that chip inherits the bug.
  • Pairing can be authenticated or not. LE Secure Connections, introduced in Bluetooth 4.2, uses elliptic curve Diffie-Hellman key exchange, and its Numeric Comparison mode gives enhanced protection against eavesdropping and man-in-the-middle attacks, according to the Bluetooth SIG. Devices with no screen or buttons often fall back to the "Just Works" mode, which has no way for the user to confirm they are pairing with the right device.
  • Addresses can be private or fixed. BLE Privacy lets a device advertise a random address that changes over time, so only trusted, paired devices can recognise it, as the Bluetooth SIG explains. The maker decides whether and how often to change it.

The Bluetooth guide in this series covers BLE pairing and stack attacks in depth.

What the attacker needs

A BLE radio within range of the target, and for stack attacks, knowledge of the specific chip. Tracking by fixed address needs only cheap receivers placed where the target goes.

Real cases

  • SweynTooth (February and July 2020). Matheus Garbelini, Sudipta Chattopadhyay and Chundong Wang of the Singapore University of Technology and Design disclosed SweynTooth, 12 vulnerabilities in a first batch and five more later, in BLE SDKs from chip vendors including Texas Instruments, NXP, Cypress, Dialog Semiconductor, Microchip, STMicroelectronics and Telink. An attacker in radio range could crash a device, deadlock its Bluetooth stack, cause buffer overflows or in some cases bypass security. The team tested a Fitbit Inspire, built on a Cypress PSoC 6 chip, and showed malicious packets could trigger a buffer overflow or temporarily deadlock its Bluetooth stack. One example record is CVE-2019-16336 in the Cypress PSoC 4 BLE component. The researchers also listed medical products that use the affected chips, including blood glucose meters and some pacemaker-related products, but said they had not verified the impact on those devices and urged makers to update their firmware.
  • Fixed BLE addresses in fitness trackers (2016). A study by Das, Pathak, Chuah and Mohapatra at UC Davis, presented at HotMobile 2016, found "majority of the fitness trackers use unchanged BLE address while advertising, making it feasible to track them." It also showed that BLE traffic volume correlated with the wearer's activity, so an eavesdropper could tell whether someone was walking, sitting or running, and that traffic patterns reflecting gait could help identify a person within a small group.

Impact

For a fitness tracker: a crash, a lost sync, or a privacy leak about where someone is and what they are doing. For a medical device that uses the same chip: potentially a crash or restart of a therapy device, which is why the SweynTooth team asked medical makers to update quickly. Note that chip vendor patches only help once each product maker ships them.

Detection

  • Users rarely see BLE attacks directly. Unexpected disconnections, failed syncs and repeated pairing prompts are worth noticing, but usually have innocent causes.
  • Makers can use BLE fuzzing in their own test labs, with authorisation, to find stack crashes before release.

Defence

  • Users: keep the tracker's firmware and the phone app updated, pair at home, and remove old pairings from your phone. Turn Bluetooth off on devices you are not using.
  • Makers: use LE Secure Connections with an authenticated pairing method wherever the device has any way to show or confirm a code; turn on BLE Privacy with regular address rotation; track the chip vendor SDK version in every firmware build; and ship vendor stack fixes as soon as they arrive. Do not rely on BLE link security alone for anything that controls therapy; add application-layer authentication.

Attack 7: How do companion apps and cloud services leak health data?

Companion apps and cloud services hold most of a wearable's data, so they leak most of it. Documented cases include an open database of 61 million fitness tracker records in 2021, a cyber attack that took Garmin Connect offline in 2020, and children's smartwatch apps that let strangers locate the wearer.

How it works and why

The watch or tracker stores a few days of data. The phone app and the cloud store years of it, along with account details, location history and links to other services. That makes them the main target and the main source of leaks. Common failure modes:

  • Misconfigured storage. A database or storage bucket left reachable from the internet without a password.
  • Weak account security. Reused passwords and no two-factor authentication.
  • Broken authorisation in APIs. The server returns another user's data when asked, because it checks that a user is logged in but not that the record belongs to them.
  • Data aggregators. Third-party services that collect data from many wearables and apps, which concentrates risk.
  • Service dependency. When the cloud goes down, sync and some features stop.

The phone security guide in this series covers the phone side of companion app risk, including malicious apps and SDKs.

What the attacker needs

Internet access and either a misconfiguration to find, a password to reuse, or a flaw in the app or API. No proximity to the wearer.

Real cases

  • GetHealth database exposure (2021). Researchers reported a non-password-protected database belonging to GetHealth, a New York company that unifies health and wellness data from wearables and apps. Website Planet's report counted 61,053,956 records, 16.71 GB, including names, dates of birth, weight, height, gender and time zone location. In a sample of more than 20,000 records, most came from Apple HealthKit and Fitbit. GetHealth secured the data the day after it was notified. It was a configuration mistake at an aggregator, not a flaw in the wearables themselves.
  • Garmin Connect outage (July 2020). Garmin said a cyber attack "encrypted some of our systems on July 23, 2020," disrupting website functions, customer support, customer-facing apps and company communications. Garmin said it had "no indication that any customer data, including payment information from Garmin Pay, was accessed, lost or stolen," and that device functionality was not affected apart from online services (BikeRadar). For several days, users could not rely on Garmin's online services.
  • Children's GPS smartwatches (2017). For the Norwegian Consumer Council's #WatchOut project, security firm mnemonic tested four children's GPS watches and found security flaws in three. The flaws could expose a child's location and activities to unauthorised parties, in two cases a third party could covertly take control of the watch, and they were "not technically difficult to exploit."

Impact

Exposure of personal and health data, loss of service, and in the worst case, a stranger able to locate a child. Health data is a special category under GDPR, which raises the legal stakes for anyone holding it.

Detection

  • Users: check the app's list of signed-in devices and connected services; watch for login alerts and password reset emails you did not request; use a breach notification service for your email address.
  • Makers: scan cloud configuration for public storage, monitor APIs for unusual access patterns such as one account reading many records, and test authorisation on every endpoint.

Defence

  • Users: a unique, long password in a password manager; two-factor authentication on the wearable account; review and remove connected apps you no longer use; for children's watches, buy from makers who publish security information and support periods.
  • Makers: never expose databases to the internet; check object-level authorisation on every API call; encrypt health data at rest; offer two-factor authentication by default; vet and monitor every data partner; and run static and dynamic testing on the app and API in every release. Our post on SBOMs for mobile applications covers inventorying app components.

Attack 8: How do wearables expose privacy when nobody hacks them?

Wearables can expose privacy by design, with no attacker involved. The clearest case is Strava's 2018 global heatmap, built from aggregated, anonymised activities, which made the outlines of sensitive facilities and the routes of people who worked there visible. Default sharing settings, public features and inferable activity data all leak information.

How it works and why

Aggregated location data is never as anonymous as it looks. A single runner's daily loop around a remote site stands out on a map even when it is merged with millions of other activities, because nobody else runs there. Activity data reveals routines: when you leave home, when you are asleep, when you are away. And default settings decide how much of this is public, because most people never change them.

What the attacker needs

Nothing technical. A public map, a public profile, or a leaderboard.

Real cases

  • Strava global heatmap (January 2018). Strava's heatmap was an aggregated view of more than a billion activities and 13 trillion GPS points. In January 2018 an analyst, Nathan Ruser, pointed out that it revealed the locations of sensitive facilities in remote regions, because the activity of the people working there stood out clearly against an empty background (Gizmodo). Strava said the map excluded activities marked private and user-defined privacy zones. In February 2018 it moved the heatmap opt-out onto the first page of its privacy settings (Engadget). Nobody broke into anything. The data was published as designed.
  • Activity inference from radio traffic (2016). As described in Attack 6, the UC Davis study showed that the volume of BLE traffic from a fitness tracker reveals whether the wearer is walking, sitting or running.

Flow from watch to Bluetooth link, companion app, maker cloud and other apps and public features, with a documented leak under each hop

Figure 8. Where wearable health data travels, with a documented leak at each hop.

Impact

Exposure of home addresses, daily routines and sensitive workplaces. For individuals, the main risks are stalking and burglary. For organisations, staff routines at sensitive sites.

Detection

Look at yourself from the outside. Log out and view your own profile and activities as a stranger would. Search the maker's public maps and leaderboards for your area.

Defence

  • Set activities to private, or followers only, by default.
  • Use privacy zones around your home and workplace.
  • Opt out of aggregated or public features such as heatmaps and leaderboards if you do not want to contribute.
  • Review which apps your wearable shares data with every few months.
  • Organisations with sensitive sites should include fitness apps in security awareness training and set clear guidance for staff.
  • Makers should make private the default, explain aggregated data use in plain language, and test aggregate releases for sparse areas where individuals stand out.

Why can't hospitals just patch medical devices?

Hospitals cannot simply patch medical devices because a device is validated as a whole, and any change can affect safety. Fixes must come from the maker, travel through the maker's supplier, testing and regulatory process, then fit a clinical downtime window. Some older devices cannot be updated at all.

Seven-step path from a published component flaw through supplier fix, maker rebuild, verification, regulatory check and hospital change to patient update, with the constraints at each step

Figure 9. A component fix has to travel through every step before it reaches a patient.

This is the part of medical device security that surprises IT teams most, so it is worth taking slowly.

The maker owns the change. EU guidance is explicit. MDCG 2019-16, the guidance on cybersecurity for medical devices endorsed by the EU's Medical Device Coordination Group, says that modifying a medical device, "e.g. the installation or enabling of third-party software including software patching, should always be under explicit published guidance of the manufacturer," and warns that unvalidated changes "can adversely affect system performance or safety in unpredictable ways." A hospital that installs a Windows update on an imaging workstation without the maker's approval may break it.

The maker depends on suppliers. If the flaw is in a licensed RTOS, network stack or BLE SDK, the maker waits for the supplier to fix it, often receiving it as a binary.

The device has to be revalidated. Medical device software is built under a defined life cycle process. MDCG 2019-16 lists IEC 62304, the medical device software life cycle standard, among the relevant standards, and MDR Annex I section 17.2 requires software to be developed to the state of the art, including verification and validation. A patch is a design change, and the maker has to test that dosing, alarms and performance still work.

The regulatory impact has to be assessed. The maker decides whether the change affects the device's conformity and updates its technical documentation.

The hospital has to schedule it. Pumps and monitors are in use around the clock. Updates need planned windows, staff time and sometimes the maker's engineers on site.

Some updates involve the patient. Abbott's 2017 pacemaker update was applied in clinic, with a very small risk of update malfunction. Medtronic's Conexus updates were applied at routine clinic visits over more than a year.

And some devices cannot be updated. Medtronic could not patch the MiniMed 508 and Paradigm pumps, so the only fix was replacement.

The practical answer is not to give up on patching. It is to make the whole path faster: makers who know every component in every build, test continuously, and keep the update path itself secure, and hospitals who compensate with segmentation and monitoring while they wait. We cover how to handle components that will never get a fix in our post on managing end-of-life software dependencies.

How can you tell if a medical device or wearable has been compromised?

For patients, the signs are device alerts, doses or settings you did not set, and unexpected pairing requests; for wearable users, unfamiliar logins and connected apps. Hospitals detect compromise through device inventories and passive network monitoring. Makers detect it through fleet telemetry, logs and vulnerability monitoring of their components.

Most of the time the honest answer for a patient is that compromise is extremely unlikely and you would see it as odd device behaviour first. Here is what to watch, by audience.

Patients with pumps and implants

  • Alerts, alarms or history entries you cannot explain.
  • Settings that have changed, doses you did not start, or a remote connection request.
  • Home monitor behaviour your clinic flags at a check-up.
  • Action: cancel unexpected doses, check your glucose or symptoms, contact the maker's support line and your clinic. Seek medical help for symptoms first.

Wearable and smartwatch users

  • Login alerts, password reset emails or two-factor prompts you did not request.
  • New devices or connected apps in your account settings.
  • Activities, routes or profile changes you did not make.
  • Action: change the password, turn on two-factor authentication, sign out all sessions, remove unknown connected apps.

Hospitals

  • New devices appearing on clinical segments, or devices talking to new destinations.
  • Internal scanning, unusual protocols, or remote access outside agreed windows.
  • Drug library or configuration changes outside the change process.
  • Action: isolate the segment, involve clinical engineering and the maker, and follow the incident plan. Clinical safety comes first.

Makers

  • Unusual programming sessions or update requests across the fleet.
  • Crash and reboot spikes after a new vulnerability is published in a component you ship.
  • Abnormal API access patterns against the cloud.
  • Action: confirm which builds contain the component, assess exploitability, communicate with customers, and ship a validated fix.

What do HIPAA and GDPR mean for health data from devices?

In plain language: HIPAA protects health data held by US healthcare providers, health plans and their business associates, so most consumer fitness trackers fall outside it. GDPR protects personal data of people in the EU, treats health data as a special category needing a strong legal basis such as explicit consent, and applies to wearables and medical devices alike.

Table showing HIPAA, GDPR, EU MDR, IEC 81001-5-1, EU CRA and EU radio equipment rules and whether each applies to a medical device and a fitness tracker

Figure 10. Which rules apply to which device. A device can fall under several at once. Not legal advice.

HIPAA in plain language

HIPAA is the US law that protects health information held by "covered entities," such as healthcare providers and health plans, and by their "business associates," the vendors that handle that data on their behalf. Whether HIPAA applies depends on who holds the data, not on what kind of data it is.

That is why the same heart rate reading can be protected in one place and not in another. A 2019 analysis by law firm Dickinson Wright puts it clearly: most fitness trackers and wellness apps are not covered by HIPAA, because they are offered directly to consumers with no link to a provider. An app a hospital develops for its patients to view records usually is covered. And once a patient asks their provider to send records to an app of the patient's choice, that copy generally leaves HIPAA's protection. By the same logic, a device or app that a provider supplies, or that a vendor runs on the provider's behalf, can fall inside HIPAA.

For device makers selling to US hospitals, that means you are often a business associate and will be asked for a business associate agreement and evidence of your safeguards. For consumers, it means reading the privacy policy of your wearable rather than assuming health privacy law covers it. Our post on HIPAA and GDPR differences goes further.

GDPR in plain language

GDPR applies to anyone processing personal data of people in the EU, whatever the device. It defines "data concerning health" as personal data about a person's physical or mental health that reveals information about their health status (Article 4). Heart rate, sleep and glucose readings tied to a person fit that description.

Article 9 makes health data a special category, which may only be processed under specific conditions, one of which is the person's explicit consent for specified purposes. In practice that means a wearable maker needs a clear legal basis for every use of health data, including sharing with partners and analytics. If health data is breached, Article 33 requires the controller to notify the supervisory authority without undue delay and, where feasible, within 72 hours, unless the breach is unlikely to put people's rights at risk.

For users, GDPR also gives rights worth using, including the right to have personal data erased when one of the grounds in Article 17 applies.

What do the EU MDR, IEC 81001-5-1 and the CRA expect from device makers?

The EU MDR requires medical device software to be developed to the state of the art, including information security, and makers to state minimum IT security requirements. IEC 81001-5-1 defines the security life cycle activities. The CRA applies to consumer wearables, not MDR devices, and requires an SBOM. EU guidance lists the SBOM among information makers should give hospitals.

The EU Medical Device Regulation (MDR)

The MDR, Regulation (EU) 2017/745, sets the general safety and performance requirements every medical device sold in the EU must meet. Two sections in Annex I matter most for software security, as summarised in MDCG 2019-16:

  • Section 17.2 requires software to be developed and manufactured "in accordance with the state of the art taking into account the principles of development life cycle, risk management, including information security, verification and validation."
  • Section 17.4 requires makers to "set out minimum requirements concerning hardware, IT networks characteristics and IT security measures, including protection against unauthorised access, necessary to run the software as intended."

MDCG 2019-16 then spells out what makers should tell hospitals. Its list of information that could be provided includes the IT security controls in the device, provisions for checking the integrity of software updates and security patches, and a Software Bill of Materials. It also expects makers to include cybersecurity in post-market surveillance, actively collecting experience "from devices on the market (including third party software and hardware components)" and acting on it in time.

IEC 81001-5-1

IEC 81001-5-1:2021, "Health software and health IT systems safety, effectiveness and security, Part 5-1: Security, Activities in the product life cycle," defines the life cycle requirements for the development and maintenance of health software needed to support conformance to IEC 62443-4-1, the secure product development standard from the industrial security family, taking the specific needs of health software into account. In plain terms, it sets out the security activities and tasks a maker should build into its software life cycle, both before release and while the product is maintained in the field. The standard stresses keeping an appropriate balance between safety, effectiveness and security, which is the same balance behind the patching constraints above.

The EU Cyber Resilience Act (CRA)

The CRA sets cybersecurity requirements for products with digital elements sold in the EU. Article 2 states that it does not apply to products covered by the MDR (2017/745) or the in vitro diagnostic regulation (2017/746). So a regulated insulin pump falls under the MDR, not the CRA.

A consumer fitness tracker or smartwatch that is not sold as a medical device is a different story. It falls under the CRA. Annex I requires makers to identify and document vulnerabilities and components, including by drawing up "a software bill of materials in a commonly used and machine-readable format," and to distribute security updates "without delay and free of charge." According to the European Commission, the CRA entered into force on 10 December 2024, reporting obligations apply from 11 September 2026, and the main obligations apply from 11 December 2027. Wireless wearables may also fall under the EU Radio Equipment Directive's cybersecurity rules, which have applied to certain categories of radio equipment since August 2025 and, according to the European Commission, are being replaced by the CRA from 11 December 2027. The smart home guide covers them in more detail.

Our posts on EU Cyber Resilience Act SBOM requirements and practical medical device SBOM requirements go deeper on both regimes.

What a good device SBOM contains

Whatever regime applies, the SBOM that hospitals and regulators find useful has the same shape:

  • Every component in the firmware, app and server software, including the RTOS, network stack, BLE SDK, cryptography and libraries, not just top-level dependencies.
  • Supplier name, component name, exact version and a unique identifier such as a package URL or CPE for each one.
  • The build or release it belongs to, so a hospital can match it to the firmware on its devices.
  • A machine-readable format such as CycloneDX or SPDX.
  • A vulnerability status for each known issue, ideally as VEX (Vulnerability Exploitability eXchange) statements saying whether the product is actually affected.
  • An update whenever the software changes, delivered through a channel the hospital can rely on.

For more detail, see our post on medical device firmware SBOMs and coordinated vulnerability disclosure.

Hardening checklists for patients, hospitals and makers

Patients should act on maker notices, keep devices updated and turn off unused remote features. Hospitals should inventory, segment and monitor every connected device and demand SBOMs in procurement. Makers should follow IEC 81001-5-1, authenticate and encrypt radios, sign updates, keep an SBOM per build and fix components fast.

Three-column hardening checklist for patients and users, hospitals and device makers

Figure 11. The checklists below, on one page.

Patients and everyday users

If you use an insulin pump, CGM, pacemaker or ICD

  • Never stop or change therapy because of a security story. Talk to your clinician.
  • Read and act on safety notices from the maker or your clinic, and ask whether your model is affected.
  • Keep follow-up appointments, where implant firmware updates are applied.
  • Keep the pump, remote and home monitor in your control and powered on as instructed.
  • Turn off remote dosing and radio features you do not use.
  • Pair devices at home, not in public.
  • Cancel any dose you did not start, check your levels, and report it.
  • Use only maker-approved apps and software with your device.
  • Keep your phone updated if it controls or reads your device.

If you use a fitness tracker or smartwatch

  • Keep the watch firmware, phone app and phone updated.
  • Use a unique password and two-factor authentication on the account.
  • Set activities to private by default and add privacy zones around home.
  • Opt out of heatmaps and public leaderboards you do not want to join.
  • Review connected apps and data sharing every few months.
  • For children's watches, choose makers with published security information and support periods.
  • Remove the device from your account and factory reset it before selling or giving it away.

Hospitals and clinics

  • Inventory every connected device with model, software and operating system version, location, owner and maker support status. Use passive monitoring to keep it current.
  • Segment by device class, deny by default, and block device-to-PC traffic. Isolate anything on an unsupported operating system.
  • Monitor device segments for new devices, scans, new destinations and unusual protocols.
  • Vendor access only through a monitored jump host, with named, time-limited accounts and recorded sessions.
  • Advisories: track maker bulletins, request SBOMs and VEX statements, and map new vulnerabilities to your inventory.
  • Procurement: require a stated support period, a vulnerability disclosure policy, an SBOM, security documentation as described in MDCG 2019-16, and a supported operating system roadmap in every contract.
  • Patching: follow the maker's approved update path on a schedule, and record and isolate exceptions.
  • Programmers and home monitors: manage them as computers, keep them updated, restrict storage, and wipe them before disposal.
  • Resilience: rehearse downtime procedures and recovery from ransomware for each clinical department.
  • Privacy: include wearables and remote monitoring in data protection impact assessments.

Device makers

  • Life cycle: follow IEC 81001-5-1 and IEC 62304, with threat modelling, secure coding, security testing and a defined process for post-release vulnerabilities.
  • Radios: authenticate both ends with per-device keys, encrypt, add replay protection, use LE Secure Connections with authenticated pairing, enable BLE Privacy, and make remote therapy features opt-in.
  • Updates: sign every firmware component, including wireless modules, verify signatures before install, and deliver updates over authenticated channels.
  • Components: keep an SBOM for every firmware build, app release and server version; monitor every component for new vulnerabilities; and require SBOMs from RTOS, SDK and module suppliers.
  • Testing: run SAST, SCA, secrets scanning and DAST on firmware applications, apps, programmers and cloud APIs in every release, plus authorised fuzzing of radio and network interfaces.
  • Cloud and apps: no public databases, object-level authorisation on every API, encryption at rest, two-factor authentication by default, and vetted data partners.
  • Disclosure: publish a vulnerability disclosure policy and advisories with affected and fixed versions, as BD and Medtronic do.
  • Documentation: give hospitals the security information MDCG 2019-16 describes, including minimum IT requirements and the SBOM.
  • Support: state a support period and plan for end of life, including what happens to cloud-dependent features.

Where does the medical device software supply chain break?

The medical device supply chain breaks where code arrives as binaries from suppliers, where one network or BLE stack ships in many products, where nobody tracks which component version is in which build, and where long validation cycles turn a supplier fix into a multi-year wait. Most vulnerable code in the cases above came from suppliers.

Look back at the cases in this guide and count how many started in someone else's code.

  • WhiteScope found more than 8,000 known vulnerabilities in 174 third-party libraries inside pacemaker programmers.
  • Ripple20 was in Treck's TCP/IP library. URGENT/11 was in IPnet, licensed into several other operating systems beyond VxWorks, which is how it reached a BD infusion pump running OSE.
  • SweynTooth was in chip vendors' BLE SDKs, shipped in fitness trackers and flagged in medical products.
  • BD's 2023 bulletin included a 2018 log4net vulnerability in a component of its own server software.
  • The GetHealth exposure was at a third-party aggregator, not at the wearable makers.

That is a software supply chain problem, with four parts.

1. Nobody knows what is inside. A maker that licenses an RTOS, a network stack and a BLE SDK as binaries may not have a full list of their components or versions. When an advisory names a component, answering "are we affected, and in which builds?" can take weeks.

2. One component, many products. Shared stacks mean one vulnerability lands on many makers and many hospitals at once. Every one of them has to do the same analysis.

3. Fixes wait on someone else, then on validation. The maker waits for the supplier, then revalidates the device, then the hospital schedules the change. Each step is there for a good safety reason, and each adds time.

4. Devices outlive their software. A scanner may be in service long after its operating system's end of support. A pump with an unfixable radio protocol may stay in use until it is replaced.

The fix is the same discipline that software companies use, adapted for safety: know every component in every build, watch them all for vulnerabilities, rank by real exposure, fix fast in your own build, revalidate efficiently, and keep the evidence for regulators and hospitals. The next section shows how Safeguard does that. For background, see our firmware supply chain security guide, SBOMs for firmware and IoT devices and software supply chain security for medical devices.

Safeguard scanning flow: firmware repositories, companion app builds, cloud services and supplier binaries go in; SBOMs, ranked vulnerabilities, fix pull requests, Gold patched packages and supplier risk and evidence come out

Figure 12. How Safeguard covers a device maker's software supply chain end to end. Raw firmware images may need extracting first.

How Safeguard helps medical device and wearable makers

Safeguard covers a device maker's software supply chain in one platform: SBOMs for firmware, apps and cloud services, binary and image scanning, reachability-ranked vulnerabilities, SCA, SAST and secrets scanning for code you own, fix pull requests gated by your CI, patched open source packages from Gold, supplier risk and compliance evidence, deployed as SaaS, in your VPC or on premises.

A medical device or wearable company usually has at least three codebases: device firmware, companion apps, and the cloud and hospital server software. Many also ship clinician programmers or home monitors. Teams often end up with a different tool for each, plus an SBOM manager, a vulnerability tracker and a GRC (governance, risk and compliance) tool for audits, each with its own inventory. Safeguard does that work end to end with one inventory and one vulnerability model, from the first SBOM to the merged fix and the evidence an auditor or notified body asks for.

See inside binaries and images you ship and receive

In the Safeguard console, Integrations > Upload App Binary accepts Windows EXE, MSI and DLL files, Android APKs, iOS IPAs, VM and disk images (QCOW2, VMDK, VHD and OVA) and container tarballs. Safeguard unpacks the file, identifies embedded libraries and their versions, maps them to known vulnerabilities with EPSS and known exploited enrichment, records licences, and creates a project with a full SBOM.

For a device maker, that covers the companion apps you ship to the app stores, the Windows software on programmers and hospital servers, supplier SDK packages, disk images for workstations, and cloud service containers. Raw firmware images may need to be extracted into their component files first, for example by unpacking the file system; Safeguard does not claim to parse every raw firmware format directly. Heavy obfuscation, encryption or unusual compression can give incomplete results, and statically linked code may show few components. More in our guide to scanning app binaries with Safeguard.

Scan the code you own

For firmware applications, apps, programmer software and cloud services in your repositories, Safeguard runs software composition analysis (SCA) across the full transitive dependency tree. Static application security testing (SAST) checks your own code for bug classes such as buffer overflows, injection and missing authentication. Secrets scanning catches API keys, signing keys and cloud credentials before they ship in an app or firmware image, which matters a great deal when a leaked signing key would let someone sign firmware. Dynamic application security testing (DAST) tests the cloud and server APIs your apps and devices call, against verified targets you own.

Rank by reachability, not by count

A programmer or hospital server can carry thousands of known vulnerabilities in its libraries, as the WhiteScope study showed. Reachability analysis shows which ones your code can actually reach, so a flaw in a network-facing parser rises above one in an unused function. For a product security team planning the next validated release, that is how a list of thousands becomes a list you can act on.

SBOMs hospitals and regulators can use

Every project produces an SBOM you can export in CycloneDX or SPDX, among many other formats, for every firmware build, app release and server version. Attestation and signing with SLSA, Sigstore and in-toto record build provenance. CSAF and VEX documents can be uploaded and linked, so the SBOM you give a hospital comes with a clear statement of which vulnerabilities actually affect the product. That is the SBOM MDCG 2019-16 lists among the information for healthcare providers, and the one the CRA requires for consumer wearables.

Fix it without waiting for upstream

This is where Safeguard differs most from a scanner. For code in your repositories, Griffin, Safeguard's remediation model, opens a fix pull request that your own CI builds and tests before anyone merges it, and your policies decide what can merge. In a regulated environment that matters: the fix enters your normal design control, verification and validation process as a reviewed change, not as an untracked edit.

For open source components where upstream has not shipped a fix, or the fix is in a version you cannot move to on a validated firmware branch, Safeguard Gold provides patched, hardened builds of those packages, and Safeguard can fork and patch the vulnerable package so your next build pulls a fixed version with the smallest possible change. For a device team, a minimal backported fix is far easier to revalidate than a major version upgrade, which can shorten the path in Figure 9.

When the vulnerable code sits inside a binary-only supplier SDK or RTOS, Safeguard tells you exactly which products and builds contain it, so you can press the supplier with evidence, publish accurate advisories and give hospitals compensating controls while you wait.

Supplier risk and compliance evidence

Safeguard's Third-Party Risk Manager lets you request SBOMs from RTOS, SDK and module suppliers and from contract manufacturers, score them and monitor them continuously (more on Safeguard TPRM). The compliance suite maps evidence from the same scans to frameworks and regulations, including HIPAA and GDPR, so SBOMs, vulnerability handling records and fix history support your MDR technical documentation, your CRA obligations for consumer products, and the security questionnaires hospitals send. Guard adds runtime guardrails on the Linux servers behind your cloud and update infrastructure, inspecting system calls, library calls and network egress.

Deploy where your code lives

Firmware source, signing keys and patient data are sensitive. Safeguard runs as SaaS in US and EU regions, in your own VPC or fully on premises, and can run in Zero AI mode with no language model in the loop, or with your own model. The local runner CLI scans on your own machines and sends only encrypted results. Safeguard is SOC 2 Type I and Type II audited.

How this compares

Medical device makers often combine a source SCA tool such as Snyk or Black Duck, a device SBOM and firmware analysis specialist such as Cybellum, Finite State or Medcrypt, and a GRC platform. Hospitals often add asset visibility platforms such as Armis or Claroty. Those are capable products, and firmware specialists go deeper on raw firmware image analysis; Armis also produced much of the public research cited here. Safeguard's case is consolidation and remediation for the maker: one inventory across source, apps, containers and binaries; reachability-ranked vulnerabilities; fix pull requests and patched packages instead of tickets; and supplier risk and compliance evidence from the same data. Safeguard can also sit alongside an existing firmware analysis tool and take over the fixing, SBOM delivery and compliance work.

The takeaway

The documented attacks on insulin pumps and pacemakers are real research, taken seriously by makers, and none has been reported used against a patient. Patients should keep using their devices, act on maker notices and turn off remote features they do not need. Wearable users get most of their protection from a strong account password, two-factor authentication and private-by-default settings.

The risks that have actually disrupted care sit elsewhere: old operating systems on flat hospital networks, cloud data left open, and shared components that put many makers' products at risk at once. Hospitals answer them with inventory, segmentation and monitoring. Makers answer them by knowing every component in every build and fixing quickly through a validated path. EU rules now expect exactly that, and the makers who can produce an accurate SBOM and a tested fix on demand will be the ones hospitals trust.

Try it yourself

Click through these interactive Safeguard demos to see the steps described in this guide.

Test yourself

Why could some older insulin pumps be controlled over the air?

Their radio protocol never checked who was sending commands. Researchers showed this on the Animas OneTouch Ping in 2016 and Medtronic MiniMed 508 and Paradigm pumps in 2019, and no confirmed attacks on patients have been reported.

What should a patient do about an insulin pump risk?

Keep using the pump, because the risk from stopping therapy is far greater than the risk of an attack. Act on any safety notice from the maker or clinic, and turn off remote features you do not use.

Why did Ripple20 and URGENT/11 reach so many devices at once?

Many medical devices use the same few embedded TCP/IP network stacks licensed from suppliers. Ripple20 was in the Treck stack and URGENT/11 in the IPnet stack, so one flaw reached products from many makers, and most makers first had to work out whether they were affected.

Why is old hospital software such an easy target?

Many devices run operating systems that no longer get security updates, because the device was validated on that version. Unit 42 found 83 percent of medical imaging devices on unsupported systems in its 2020 report, and malware on a flat hospital network can spread to them.

Why can't hospitals just patch medical devices?

A device is validated as a whole, so a change can affect safety. The fix must come from the maker, pass through supplier, testing and regulatory steps, and fit a clinical downtime window. Some devices, such as the MiniMed 508 and Paradigm pumps, could not be patched at all.

What did the Strava 2018 heatmap show about wearables?

Wearables can expose privacy by design, with no attacker involved. Aggregated, anonymised activities made the outlines of sensitive facilities and the routes of people who worked there visible.

Frequently asked questions

Can an insulin pump or pacemaker be hacked?

Researchers have shown that some older insulin pumps and some implant telemetry and programmer systems could be accessed by a skilled person nearby with specialist equipment. Makers responded with firmware updates, disabled update paths and replacement programmes. Attacks need proximity and deep expertise. Keep using your device, keep follow-up appointments and act on maker safety notices.

Has anyone been harmed by a hacked medical device?

In the cases covered here, Medtronic and BD reported no confirmed exploitation of the vulnerabilities, and Medtronic reported no patient harm. Real disruption to care has come mostly from ransomware on hospital systems, such as WannaCry in 2017, which locked 34 NHS trusts out of systems and devices such as MRI scanners and cancelled appointments.

Why do hospitals still run old Windows on medical equipment?

Because the device was tested and approved as a whole on that operating system. EU guidance says patches must follow the maker's published instructions, since unvalidated changes can affect safety. The maker must test and release an update first, and some older devices never get one. Hospitals compensate by isolating these devices on their own network segments.

Can my fitness tracker or smartwatch be hacked?

Yes, though serious attacks are uncommon. Researchers have crashed trackers through Bluetooth stack flaws and tracked them through fixed radio addresses. The bigger risks are your account and the maker's cloud: reused passwords, leaky connected apps and exposed databases. Keep the device and app updated, use two-factor authentication and set activities to private.

Is the health data from my fitness tracker protected by HIPAA?

Usually not. HIPAA covers health data held by US healthcare providers, health plans and their business associates. A tracker you bought yourself and use with the maker's app normally falls outside it, so the maker's privacy policy governs your data. A monitoring device your clinic prescribes and runs generally falls inside HIPAA through the provider.

What does GDPR say about health data from wearables?

GDPR treats health data as a special category. A wearable maker processing heart rate, sleep or glucose data of people in the EU needs a specific legal basis, such as explicit consent, for each purpose. It must notify the regulator of a breach within 72 hours where feasible, and users can access, correct and delete their data.

Do medical device makers have to provide an SBOM?

EU guidance MDCG 2019-16 lists a software bill of materials among the information makers should provide to hospitals, and expects post-market surveillance to cover third-party software components. For consumer wearables outside the MDR, the EU Cyber Resilience Act requires a machine-readable SBOM from 11 December 2027. Hospitals increasingly ask for one in contracts.

Does the EU Cyber Resilience Act apply to medical devices?

No. CRA Article 2 excludes products covered by the Medical Device Regulation and the In Vitro Diagnostic Regulation, which have their own cybersecurity requirements. Consumer fitness trackers and smartwatches that are not medical devices do fall under the CRA, with reporting duties from September 2026 and full requirements from December 2027.

Sources

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

We email you a link to confirm, and only then add you to the list. Unsubscribe in one click. Privacy

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.