Safeguard
Supply Chain Security•

How Hospitals Get Hacked and How to Secure Them

How hospitals get hacked: EHR, PACS, HL7 and FHIR, legacy Windows, vendor access and ransomware. Real cases and a hardening plan for health IT.

Safeguard Security Team
67 min read

Hospitals rarely get hacked through a clever trick against one machine. They get hacked through a stolen password on a remote access portal without multi-factor authentication, an unpatched Windows server, an imaging archive left open to the internet, or a supplier whose outage becomes the hospital's outage. Ransomware then turns that foothold into a downtime that clinicians have to work through on paper. Every step in that chain has a known defence, and this guide walks through them one at a time.

A modern hospital is a dense software system. An electronic health record sits at the centre. Around it run imaging archives, laboratory and pharmacy systems, interface engines that pass thousands of messages an hour, thousands of networked clinical devices, identity services, and a long list of vendors with remote access. Then come the outside services the hospital depends on for claims, pathology and prescriptions. When one of those parts fails, care slows down. When several fail at once, care stops in places.

This guide is for three groups of readers. Hospital and clinic IT, security and clinical engineering teams who defend that system day to day. Health IT vendors who build the software hospitals run, and who will find the supply chain, SBOM and secure development sections most useful. And board members, clinicians and managers who want to understand what actually goes wrong and what good looks like. Jargon is defined the first time it appears.

It 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. Incidents are described technically, by their business and patient impact, without attributing them to any group. A note on the law: testing systems you do not own, or do not have written permission to test, is illegal in most countries. In a hospital it can also put patients at risk. Products and research teams are named only as context for authorised work.

Individual medical devices and wearables (insulin pumps, pacemakers, infusion pumps, fitness trackers) have their own guide: how medical devices and wearables get hacked. This post links to it rather than repeating it, and treats devices here as a fleet the hospital has to manage. Other guides in the series cover how laptops get hacked, how routers get hacked, how printers get hacked and how Wi-Fi gets hacked, all of which are present in every hospital. For where the hardware and firmware come from, see how a laptop is built and fabless chip designers and the software supply chain.

Contents

  1. How do hospitals get hacked? The short answer
  2. How does a hospital's IT actually work?
  3. What is the attack surface of a hospital?
  4. Which hospital attacks matter most in practice?
  5. Attack 1: How are EHR and EMR platforms attacked?
  6. Attack 2: Why are PACS and DICOM imaging systems exposed?
  7. Attack 3: How are HL7 and FHIR interfaces and APIs attacked?
  8. Attack 4: What happens when lab and pharmacy systems are hit?
  9. Attack 5: Why are clinical networks and legacy Windows so exposed?
  10. Attack 6: Why are biomedical device fleets hard to defend?
  11. Attack 7: How do attackers get in through vendors and remote access?
  12. Attack 8: How do identity theft and phishing open the door?
  13. How does ransomware unfold inside a hospital?
  14. What did WannaCry, Change Healthcare and Synnovis teach hospitals?
  15. What keeps patients safe when systems go down?
  16. How can a hospital detect an attack early?
  17. What do HIPAA and GDPR mean for a hospital breach?
  18. Where does the healthcare software supply chain break?
  19. Hardening checklists for hospitals, clinics and health IT vendors
  20. How Safeguard helps hospitals and health IT vendors
  21. The takeaway
  22. Frequently asked questions
  23. Sources

How do hospitals get hacked? The short answer

Hospitals get hacked mostly through ordinary IT weaknesses: stolen or phished credentials on remote access without multi-factor authentication, unpatched internet-facing systems and old Windows, imaging and web systems exposed to the internet, flat clinical networks, and suppliers whose compromise or outage spreads into the hospital. Ransomware then encrypts systems and steals data.

Three points set the tone for the rest of this guide.

First, the damage that matters most is to care, not only to data. When the WannaCry ransomware spread in May 2017, 34 NHS hospital trusts in England were infected and locked out of systems and devices such as MRI scanners, according to a 2019 study in npj Digital Medicine. Five acute trusts diverted ambulances. In 2024, a pathology supplier's outage in London disrupted more than 10,000 appointments, and a later safety investigation found a long wait for a blood test result was one of the factors in a patient's death.

Second, the hospital's perimeter now includes its suppliers. The 2024 Change Healthcare attack did not touch most hospitals' own networks, yet 94% of US hospitals surveyed by the American Hospital Association reported a financial impact and 74% reported an impact on direct patient care.

Third, almost every documented case started with something basic. A portal without multi-factor authentication. A Windows patch that had been available for two months. A database answering anyone on the internet. That is good news. Basic weaknesses have well understood fixes, and the hardening sections below put them in order.

How does a hospital's IT actually work?

A hospital runs an electronic health record at its centre, connected through an interface engine to imaging, laboratory, pharmacy, scheduling and billing systems, and to thousands of networked clinical devices. Messages move as HL7 v2, FHIR and DICOM. Identity comes from a central directory, and many systems are maintained remotely by vendors.

Map of a hospital's systems: the EHR at the centre, connected through an interface engine to PACS, laboratory, pharmacy, billing and clearinghouse, with clinical device networks, identity, vendor remote access and outside suppliers around it

Figure 1. The hospital as a system: one record at the centre, dozens of systems and suppliers around it.

It helps to know the main parts by name, because each one is attacked differently.

The electronic health record (EHR). An EHR is the hospital's main clinical system: patient demographics, notes, orders, results, medications, allergies and billing codes. EMR (electronic medical record) is an older term often used for a single practice's record. Large hospitals usually run one major commercial EHR; smaller clinics may run a hosted or open source product. The EHR is a large web and database application with a client, an application tier, a database and a reporting copy.

The interface engine. Hospitals have dozens of systems that must exchange data. An interface engine sits in the middle and routes messages: an admission from the EHR to the lab, a result from the lab back to the EHR, an order to radiology. Most of this traffic uses HL7 version 2, a decades-old text message format with segments separated by line breaks, carried over a thin framing layer called MLLP (Minimal Lower Layer Protocol). Newer integrations use FHIR (Fast Healthcare Interoperability Resources), which exposes health data as web APIs in JSON.

PACS and imaging. A PACS (picture archiving and communication system) stores and serves medical images from CT, MRI, X-ray and ultrasound machines, which are called modalities. Images move in DICOM (Digital Imaging and Communications in Medicine), a standard that covers both the file format and the network protocol. Radiologists read images on workstations, often remotely.

Laboratory and pharmacy systems. A laboratory information system (LIS) receives test orders, tracks samples through analysers and returns results. Blood banks use their own systems to match donor blood to patients. Pharmacy systems manage the formulary, verify orders and drive automated dispensing cabinets on the wards.

Clinical devices. Patient monitors, infusion pumps, ventilators, imaging modalities, lab analysers and many more connect to the network. Clinical engineering (also called biomedical engineering) maintains them.

Identity and the office stack. Almost every hospital runs a central directory such as Microsoft Active Directory for logins, plus email, file shares and remote access for staff and vendors.

Outside services. Claims go to payers through a clearinghouse, a company that checks and forwards billing transactions. Pathology may be outsourced. E-prescriptions go through national networks. Each of these is a dependency the hospital does not control.

Data flow diagram: an order travels from the EHR through the interface engine as HL7 v2 to the laboratory system, the result returns, images move by DICOM from modality to PACS, and claims leave through a clearinghouse

Figure 2. How data moves: HL7 v2 between internal systems, DICOM for images, FHIR APIs for apps, and claims through a clearinghouse.

The software inside all of this comes from many places. The EHR vendor ships its own code plus hundreds of open source libraries. The PACS vendor may ship a Windows server image. The lab analyser runs an embedded Windows build the manufacturer validated years ago. That layering is why the supply chain section near the end matters as much as the network sections.

What is the attack surface of a hospital?

A hospital's attack surface is everything an attacker can reach: internet-facing portals, remote access and email, exposed imaging and web services, internal interfaces between clinical systems, legacy servers and devices on shared networks, vendor connections, and the suppliers whose systems the hospital relies on.

Entry pointWhat it isWhat an attacker needs
Remote access portals and VPNsHow staff and vendors log in from outsideA valid password, or an unpatched gateway
EmailEvery staff mailboxOne person to open a link or attachment, or reuse a password
Patient and clinician portalsWeb apps on the internetA web application vulnerability or stolen patient credentials
FHIR and partner APIsData access for apps and partnersWeak authorisation checks, or leaked API keys
PACS and DICOM servicesImage archives and viewersAn archive reachable from the internet or a flat network
Interface engine and HL7 feedsInternal message routingA foothold inside the clinical network
Legacy servers and workstationsOld Windows that cannot be patchedNetwork access and a known vulnerability
Clinical devicesMonitors, pumps, modalities, analysersNetwork access; see the device guide
Vendor connectionsRemote support tools and service accountsCompromise of the vendor or its credentials
Outside suppliersClearinghouses, pathology, hosted EHRsNothing at the hospital at all: the supplier is the target

The last row is the one many plans miss. In two of the three ransomware cases below, the hospitals affected were not the organisations attacked.

Which hospital attacks matter most in practice?

In practice, the attacks that do the most harm are the cheap, broad ones: credential theft on remote access, unpatched internet-facing systems and ransomware spreading through flat networks, and outages at shared suppliers. Targeted attacks on a single device are rare by comparison and need far more effort.

Matrix placing hospital attack classes by attacker effort and number of patients affected, with supplier ransomware and credential theft in the high-impact, low-effort corner

Figure 3. An editorial view of where to start: broad and cheap attacks first.

A long-running data set backs this up. A 2022 study in JAMA Health Forum counted 374 ransomware attacks on US health care delivery organisations from January 2016 to December 2021. They exposed the personal health information of nearly 42 million patients, and the annual count more than doubled from 43 to 91 over the period. Almost half of the attacks (44.4%) disrupted the delivery of care, most often through electronic system downtime, and some led to cancelled care and ambulance diversion (Neprash et al., via Europe PMC).

Timeline of hospital cyber incidents and vulnerabilities from 2017 to 2026, including WannaCry, the Allscripts outage, the OpenEMR report, exposed PACS research, BlueKeep, the Romanian hospital software attack, Change Healthcare and Synnovis

Figure 4. Ten years of hospital security events, each documented by a source cited in this article.

Attack 1: How are EHR and EMR platforms attacked?

EHR platforms are attacked through their web portals, their application and database servers, the accounts that use them, and the hosting providers that run them. Because the EHR holds every patient's record and drives every order, both data theft and an outage are serious, and an outage hits every ward at once.

How it works and why

An EHR is a large web application with many users, many roles and a long history. That brings the familiar web flaws: broken authentication, injection into database queries, file upload bugs and missing access checks on individual records. Patient portals put part of the application on the internet for millions of users. And the EHR's own servers run on Windows or Linux with all the patching that implies.

The bigger risk for most hospitals is not a flaw in the EHR code. It is everything around it: the accounts with access, the servers it runs on, the directory it trusts, and, when it is hosted, the vendor's data centre.

What the attacker needs

For a web flaw, network access to the portal and a vulnerability in the version deployed. For the far more common route, valid credentials or a foothold elsewhere in the network from which to reach the EHR servers. For a hosted EHR, a way into the hosting provider, which then affects every customer at once.

Real cases

OpenEMR, 2018. OpenEMR is a widely used open source EHR and practice management system. In 2018 a volunteer research group, Project Insecurity, published a report on version 5.0.1.3 describing a patient portal authentication bypass, multiple SQL injection flaws, several remote code execution issues, unauthenticated administrative actions and arbitrary file read and write. The disclosure was handled well: the researchers contacted the project on 7 July 2018 and the project pushed fixes on 20 July, before public release in August. The portal bypass is tracked as CVE-2018-15152, which let a remote attacker reach portal pages such as lab results, medications and patient documents in versions before 5.0.1.4.

A hosted EHR outage, 2018. In January 2018, ransomware infected the Raleigh and Charlotte data centres of Allscripts, an EHR and e-prescribing vendor that, according to HIPAA Journal, served 2,500 hospitals and 19,000 post-acute care organisations. Several applications went offline for up to 1,500 clients. E-prescribing came back within days; some practices waited longer for their EHR. No hospital network was breached, yet clinicians could not reach their records.

A national hospital system, 2024. In February 2024, attackers breached the Romanian software company behind Hippocrates, a hospital information system used for admissions, lab tests, pharmacy logistics and payroll. Ransomware spread to hospitals running it. The BBC reported that more than 100 hospitals were told to disconnect from the internet, 26 were found to be infected, and staff switched to pen and paper. That case is covered again under downtime.

Impact

Data theft from an EHR exposes the most sensitive records a person has. An EHR outage removes order entry, results, medication records and bed management at the same moment, which is why every hospital needs a tested downtime plan.

Detection

Watch for unusual record access patterns: a single account opening thousands of charts, access outside a user's normal department or hours, and bulk exports or reports. Most EHRs keep detailed access logs; send them to your security monitoring, not only to the privacy office. On the servers, watch for new administrative accounts, new scheduled tasks and unexpected outbound connections from database hosts.

Defence

  • Keep the EHR and its database, web and reporting tiers on supported, patched versions, following the vendor's tested release notes.
  • Put patient portals behind a web application firewall and test them with authenticated dynamic testing (DAST) on systems you own.
  • Enforce multi-factor authentication for every remote and privileged EHR account, and for all portal administrators.
  • Limit who can run bulk reports and exports, and review those rights quarterly.
  • For hosted EHRs, ask the vendor for its incident response plan, recovery time objectives, SOC 2 report and software bill of materials (SBOM), and test your own downtime plan against a full vendor outage.

Attack 2: Why are PACS and DICOM imaging systems exposed?

PACS and DICOM systems are exposed because many imaging archives are reachable from the internet with no authentication, so anyone can query them for patient names and scans. Researchers have found thousands of such servers every time they looked, in 2019, 2023 and again in late 2025.

How it works and why

DICOM servers answer network requests to find, retrieve and store images. Each node is identified by an Application Entity Title (AE Title), a short name. Many products accept any caller that presents a plausible AE Title, and vendors publish their default titles in their documentation. DICOM does support TLS encryption, but it is rarely switched on. When an archive is connected straight to the internet, often to let a remote radiologist or a referring clinic see images, anyone who finds it can ask it for studies.

There is a second, quieter issue in the file format. The DICOM standard reserves a 128-byte preamble at the start of every file. CVE-2019-11687 records that this preamble can hold executable headers for Windows (PE) or Linux (ELF), which means a file can be a valid medical image and a valid program at the same time. Malware could hide inside files that look like patient images and that teams are reluctant to scan or delete.

PACS software also has ordinary bugs. CVE-2023-33466 in Orthanc, a popular open source DICOM server, let authenticated API users overwrite arbitrary files, which in some deployments could lead to remote code execution. It was fixed in Orthanc 1.12.0.

What the attacker needs

For an exposed archive, nothing more than an internet connection, the server's address (which internet scanning services list) and a DICOM client. That is exactly why it matters. Tampering with images needs write access, which some exposed servers also allow.

Real cases

2019. A joint investigation by ProPublica and the German broadcaster Bayerischer Rundfunk, building on work by Greenbone Networks, identified 187 servers in the US that made patient records available without a password. Records covered more than 5 million US patients, and medical data from more than 16 million scans worldwide was available online, including names, birth dates and in some cases Social Security numbers. Greenbone found problems in at least 52 countries.

2023. At Black Hat Europe, the German consultancy Aplite reported more than 3,800 exposed DICOM servers in over 110 countries, exposing personal information of some 16 million patients and more than 43 million health records. More than 70% of the exposed servers were hosted by cloud providers. Aplite said fewer than 1% of DICOM servers on the internet used effective security measures, and that 39 million records were at risk of tampering. The Medical Imaging and Technology Alliance, which oversees DICOM, responded that the standard "does not inherently pose a security risk" and that security is a shared responsibility between device makers and health providers. Both statements are true at once: the protections exist and are not used.

Late 2025. Trend Micro scanned again between November and December 2025 and published its results in May 2026: 3,627 DICOM servers across more than 100 countries reachable from the public internet. Only 0.14% used TLS, and 99.56% accepted connections without AE Title validation. The US accounted for about a third.

Impact

Exposure of names, dates of birth, referring doctors and the images themselves. Images are permanent: unlike a password, a CT scan cannot be changed after it leaks. Write access adds the risk of altered images, which is a patient safety problem, not only a privacy one.

Detection

Scan your own public address ranges for DICOM services on the usual ports (Aplite lists TCP 104, 11112 and 4242) and for web viewers. Check internet search engines for your organisation's ranges. Inside, alert on DICOM associations from any address that is not a known modality, workstation or partner, and on large query and retrieve volumes.

Defence

Aplite's recommendations are a sound order of work:

  1. Remove public exposure. No DICOM service on the internet. Connect remote sites and hosted archives through a VPN or IPsec tunnel.
  2. Segment imaging. Put modalities and the archive in a dedicated network zone, and allow the DICOM protocol only from modalities.
  3. Control access. Allow only known modality addresses, use non-default AE Titles where AE Title checks are all the product supports, and give people access through DICOMweb (DICOM over HTTPS) tied to your identity provider, behind a web application firewall with TLS.

Then keep PACS servers patched, enable TLS where the product supports it, scan stored DICOM files for embedded executables, and back up the archive offline.

Attack 3: How are HL7 and FHIR interfaces and APIs attacked?

HL7 v2 feeds are attacked from inside the network, where an attacker can read or inject messages that systems trust by default. FHIR APIs are attacked from outside through weak authorisation, where an app or user can request records that are not theirs. FHIR itself defines no security; it relies on TLS, OAuth and correct access checks.

How it works and why

HL7 v2 messages travel between systems inside the hospital, traditionally over MLLP. MLLP is a minimal framing layer. The HL7 over HTTP specification was written partly because HTTP lets authentication be specified "in a standardized way", which tells you that classic MLLP feeds leave authentication and encryption to whatever surrounds them: TLS wrappers, VPNs and network segmentation. Receiving systems tend to trust any well formed message on their interface port. An attacker inside the clinical network who can reach those ports can read patient data as it passes and, in principle, inject false admissions, orders or results.

FHIR is different. It is a modern web API, so it inherits web API risks. HL7's own FHIR security page is blunt: "FHIR is not a security protocol, nor does it define any security related functionality." It recommends TLS for all production traffic, authenticated clients, OAuth for web-centric use and the SMART App Launch framework, which builds on OAuth 2.0. The most common API flaw is what OWASP calls broken object level authorisation: the server checks that the caller is logged in but not that the specific record requested belongs to them. On a FHIR server, that means changing a patient ID in a request and getting someone else's record.

What the attacker needs

For HL7 v2, a foothold on the clinical network with a route to the interface engine or a receiving system. For FHIR, an account or app registration (often available to any patient or developer) plus an implementation that skips per-record checks, or leaked API keys and client secrets.

Real cases

Public breach reports rarely name the interface as the root cause, so the strongest evidence here is structural rather than a single incident. The OpenEMR report above includes patient portal pages that returned lab results, medications and documents without proper authentication, which is the same class of flaw. The exposed DICOM servers in Attack 2 show what happens when a clinical protocol designed for trusted networks is put on the internet. FHIR moves more health data onto internet-facing APIs, so the authorisation discipline that web teams apply to banking APIs now applies to health records.

Impact

Silent data theft at scale through an API, or quiet corruption of clinical data through injected messages. A false result or a false allergy entry is a safety issue, and it is hard to spot because it arrives through a trusted channel.

Detection

  • Inventory every interface: source, destination, port, message types and owner. Most hospitals discover forgotten feeds when they do this.
  • Alert on new connections to interface engine ports from hosts that are not known senders.
  • For FHIR, log every request with the authenticated user, client and patient ID, and alert when one client reads many patients' records, or when a patient-facing app requests records for patients other than the logged-in user.

Defence

  • Wrap HL7 v2 connections in TLS or carry them over VPN tunnels between sites, and restrict interface ports to known senders with firewall rules.
  • Validate messages at the interface engine: reject unexpected message types, unknown sending facilities and malformed segments.
  • For FHIR, use SMART App Launch with narrow scopes, enforce per-record authorisation on the server for every request, rate limit, and test the API for broken object level authorisation before every release.
  • Keep API keys and client secrets out of code and mobile apps, and rotate them.

Attack 4: What happens when lab and pharmacy systems are hit?

When laboratory systems go down, blood tests, transfusion matching and results slow to a crawl, which delays surgery, cancer care and emergency decisions. Pharmacy systems and dispensing cabinets hold privileged accounts and drug data. Both are high-value targets because care depends on them within minutes.

How it works and why

A laboratory information system talks to dozens of analysers, each a computer with its own operating system and network connection, and to the EHR through the interface engine. Many hospitals outsource pathology to a partner, which puts the lab on someone else's network. Pharmacy systems drive automated dispensing cabinets on the wards, which run Windows and connect to central servers.

What the attacker needs

For a lab attack, a foothold in the lab provider's network or the hospital network segment where the LIS and analysers sit. For dispensing systems, network access and credentials, and default or shared credentials make that easier.

Real cases

Synnovis, 2024. Synnovis is a pathology partnership between Guy's and St Thomas' NHS Foundation Trust, King's College Hospital NHS Foundation Trust and the lab company SYNLAB. On 3 June 2024 it suffered a ransomware attack that, in its own words, "impacted almost all Synnovis IT systems" (Synnovis incident updates). The full case is told below in the ransomware section.

Dispensing cabinets. CVE-2022-22767 records that specific BD Pyxis products, which are automated medication dispensing systems, were installed with default credentials and might still be running with them. Some installations shared the same local operating system or domain credentials across product types. The CVE record warns that someone with those credentials may be able to gain privileged access to the underlying system. The lesson is about default and shared credentials on clinical systems in general.

Impact

For a lab: delayed results, cancelled operations, a shift to slower manual processes and a strain on blood supply. For pharmacy: disrupted medication supply on the wards and exposure of controlled drug records.

Detection

Monitor the LIS, blood bank and dispensing servers like the critical systems they are: endpoint detection, privileged account monitoring and alerting on new remote sessions. Track turnaround times. A sudden drop in result volume can be an early sign of a problem.

Defence

  • Change every default credential at installation and replace shared accounts with named or managed ones. Ask vendors to confirm in writing that no default credentials remain.
  • Put analysers, the LIS and dispensing servers in their own network zones with tight conduits to the EHR and interface engine.
  • Treat an outsourced lab as part of your critical system: agree recovery times, see its security evidence, and plan for its total loss.
  • Keep paper request and result forms, manual blood bank procedures and a supply of emergency blood planning in the downtime plan.

Attack 5: Why are clinical networks and legacy Windows so exposed?

Clinical networks are exposed because many hospitals run medical and imaging systems on Windows versions that no longer get updates, often on flat networks where any infected computer can reach them. Old Windows carries well known, wormable vulnerabilities that ransomware uses to spread in minutes.

How it works and why

A device or clinical workstation is validated as a whole. Its maker tested it on a particular operating system, so the hospital usually cannot upgrade that operating system without the maker's support. The result is a long tail of old Windows: Unit 42's 2020 IoT Threat Report found that 83% of medical imaging devices it observed ran unsupported operating systems, a jump driven by the end of Windows 7 support, and that 98% of IoT device traffic was unencrypted.

Old Windows matters because of a few famous flaws:

  • EternalBlue, MS17-010. A flaw in the SMBv1 file sharing service (CVE-2017-0144 and related CVEs) that let an attacker run code by sending crafted packets. Microsoft published the fix in bulletin MS17-010 on 14 March 2017. WannaCry used it two months later.
  • BlueKeep. CVE-2019-0708 in Remote Desktop Services let an unauthenticated attacker run code by connecting over RDP. Any old Windows system with RDP exposed is at risk.
  • Zerologon. CVE-2020-1472 in the Netlogon protocol let an attacker on the network take over a domain controller. Once the directory falls, every system that trusts it falls with it.

A flat network, where workstations, servers and devices share one large segment, lets a worm or an operator reach every one of those flaws from any infected laptop.

What the attacker needs

A foothold anywhere on the internal network, from a phishing email or stolen remote access, and a route to vulnerable hosts. Worms need nothing more.

Real cases

WannaCry in 2017 is the defining case, covered in full below. Reporting on the official UK review of the incident said NHS trusts had not acted on critical alerts to patch or move away from vulnerable older software, and that better firewall management could also have helped.

Impact

Fast, wide outages. A worm on a flat network can encrypt clinical workstations, imaging systems and servers in hours.

Detection

Keep a live inventory that records the operating system and support status of every host, including devices. Watch for SMBv1 and RDP traffic between segments that should not talk, scanning behaviour from workstations, and new connections to domain controllers.

Defence

  • Patch what can be patched, quickly, starting with internet-facing systems and anything listed in known exploited vulnerability catalogues.
  • For what cannot be patched, isolate: place each legacy system in a zone with only the conduits it needs, disable SMBv1 and RDP where the maker allows, and block them between zones regardless.
  • Protect domain controllers as the most critical systems in the hospital, with tiered administration and separate admin accounts.
  • Plan replacement. Make supported operating systems and a patch path a purchasing requirement for every new clinical system.

Attack 6: Why are biomedical device fleets hard to defend?

Biomedical device fleets are hard to defend because a large hospital has thousands of connected devices from hundreds of makers, many with old software, long lives and patching that only the maker can do. The practical defence is inventory, segmentation and maker coordination, not patching every device yourself.

The details of how pumps, monitors and implants are attacked are in our medical devices and wearables guide, including shared network stack flaws, infusion pump vulnerabilities and the regulations makers work under. For a hospital, the fleet view is what counts:

  • Know what you have. An accurate inventory of every networked device, with make, model, software version, location, owner and network address. Clinical engineering and IT need to share one list.
  • Know what is inside. Ask makers for an SBOM and their security documentation at purchase, so that when a library vulnerability is announced you can tell which devices carry it.
  • Segment by function. Put devices in zones by type and risk, allow only the conduits they need (to the EHR, the monitoring central station or the maker's update server), and block everything else.
  • Manage changes through the maker. Apply patches the maker has validated, on the maker's instructions. IEC 80001-1:2021, the international standard for risk management of IT networks with medical devices, sets out how a healthcare organisation should manage safety, effectiveness and security when connecting devices (IEC).

Attack 7: How do attackers get in through vendors and remote access?

Attackers get in through vendors by stealing or guessing the credentials vendors use for remote support, by exploiting flaws in remote access gateways and support tools, or by attacking a supplier directly so that its outage becomes the hospital's outage. Change Healthcare began with a remote access portal without multi-factor authentication.

How it works and why

Hospitals rely on vendors to maintain their EHR, PACS, lab systems and devices, and many vendors connect in remotely. Each connection is a door: a VPN account, a remote desktop gateway, a remote support tool or a service account with broad rights. These doors are often older than the current security team, shared between vendor staff and exempt from multi-factor authentication "because the vendor's tool does not support it".

Remote access products are also among the most attacked software on the internet. Three examples from the CVE record:

  • CVE-2023-4966 in NetScaler ADC and Gateway, an unauthenticated information disclosure in gateway configurations that could expose session data.
  • CVE-2024-1709, an authentication bypass in ConnectWise ScreenConnect 23.9.7 and earlier, a remote support tool used by many IT providers.
  • CVE-2023-34362, a SQL injection in Progress MOVEit Transfer, a file transfer product widely used to exchange data with partners, which let an unauthenticated attacker reach its database.

None of these is specific to healthcare. That is the point. Hospitals run the same edge products as everyone else, and attackers scan for them everywhere.

What the attacker needs

Valid vendor credentials (bought, phished or reused from another breach), or an unpatched gateway. For a supply chain outage, nothing at the hospital at all.

Real cases

Change Healthcare, 2024. In public testimony in May 2024, UnitedHealth Group's chief executive said attackers used compromised credentials on 12 February 2024 to access a portal for remote access to desktops, and that the portal did not have multi-factor authentication turned on. They deployed ransomware nine days later (Healthcare Dive). The full case follows in the ransomware section.

Synnovis and Hippocrates, 2024. In both, the organisation attacked was a supplier: a pathology provider in London and a hospital software company in Romania. Hospitals that had done nothing wrong lost services.

Impact

A single vendor account can open many hospitals. A single supplier outage can stop claims, lab work or records for an entire region.

Detection

Log and review every vendor session: who, when, from where, to which system. Alert on vendor logins outside agreed windows, from new countries or devices, or to systems outside the vendor's contract. Watch for new remote support tools installed on servers; attackers often bring their own.

Defence

  • Route all vendor access through one managed gateway with multi-factor authentication, individual named accounts, session recording and approval for each session.
  • Remove standing access. Enable vendor accounts only for the window of the work.
  • Patch remote access gateways and support tools as an emergency when vulnerabilities in them are announced, and subscribe to their advisories.
  • Write security into contracts: multi-factor authentication, breach notification times, SBOMs, vulnerability handling, recovery times and the right to audit.
  • Map which suppliers your clinical services depend on, and write downtime plans for losing each critical one, including clearinghouses and outsourced labs.

Attack 8: How do identity theft and phishing open the door?

Phishing and credential theft give attackers a real user's access, which is why most hospital intrusions start with a login rather than an exploit. Multi-factor authentication on every remote and privileged login blocks most of these attempts, and Microsoft reports it can block over 99.9% of account compromise attacks.

How it works and why

A hospital has thousands of staff, many on shared workstations, many rotating between sites, and all receiving email from patients, suppliers and other hospitals. Phishing emails steal passwords through fake login pages or deliver malware through attachments. Passwords reused from other breaches get tried against remote access portals. Once inside, attackers target the directory (Active Directory), because domain administrator rights unlock everything.

What the attacker needs

An email address list and a convincing message, or a list of leaked passwords. Then a login page that accepts a password alone.

Real cases

Change Healthcare's intrusion began with compromised credentials on a portal without multi-factor authentication, as described above. Microsoft's analysis of account attacks found that multi-factor authentication can block over 99.9% of account compromise attacks, and warned that legacy protocols which cannot handle multi-factor authentication give attackers a way around it.

Impact

Full user access to email, files and clinical systems, and a path to domain administrator rights.

Detection

Alert on impossible travel, logins from new devices or countries, multi-factor prompts the user did not start, new mailbox forwarding rules, and service accounts logging in interactively. Watch domain controllers for new privileged group members.

Defence

  • Multi-factor authentication on every remote access path, email, cloud admin console and privileged account. Prefer phishing-resistant methods such as FIDO2 security keys or platform passkeys for administrators.
  • Block legacy authentication protocols that cannot do multi-factor authentication.
  • Fast, low-friction login for shared clinical workstations (badge tap plus PIN), so that staff do not share sessions to save time.
  • Separate admin accounts, tiered administration and just-in-time privileged access.
  • Short, practical phishing training with a one-click "report this" button, and thanks rather than blame for reports.

How does ransomware unfold inside a hospital?

Hospital ransomware follows a predictable path: initial access through stolen credentials, phishing or an unpatched edge system; days of quiet movement and privilege escalation; theft of data; then encryption of servers and backups. The days between entry and encryption are the defender's best chance to stop it.

Diagram of a ransomware attack in five stages with what defenders can detect at each stage, using the nine days between initial access and encryption at Change Healthcare as an example

Figure 5. The anatomy of a hospital ransomware attack, and where defenders can catch it.

The stages, described for defenders:

  1. Initial access. A stolen password on remote access, a phishing email, or an unpatched gateway.
  2. Discovery and privilege escalation. The attacker maps the network, finds the directory and backup servers, and works towards domain administrator rights, often using legitimate admin tools so that activity blends in.
  3. Data theft. Large volumes of files are copied out, so the attacker can threaten to publish them.
  4. Destroying recovery. Backups, snapshots and recovery consoles are deleted or encrypted.
  5. Encryption. Servers and workstations are encrypted, usually at night or at a weekend, and a ransom note appears.

At Change Healthcare, nine days passed between initial access and encryption. At many hospitals the gap is similar, measured in days, not minutes. Every stage before the fifth leaves traces: new admin accounts, remote sessions to domain controllers, backup deletions, unusual outbound data volumes. That is what monitoring is for.

What did WannaCry, Change Healthcare and Synnovis teach hospitals?

WannaCry showed how unpatched Windows on flat networks turns a worm into a national outage. Change Healthcare showed how one supplier's missing multi-factor authentication can freeze payments and prescriptions across a country. Synnovis showed how a pathology supplier's outage harms patients for months.

Three case cards comparing WannaCry 2017, Change Healthcare 2024 and Synnovis 2024 by entry point, what failed, the scale of disruption and the main lesson

Figure 6. Three documented cases side by side, described technically and by impact.

WannaCry and the NHS, May 2017

What happened. On Friday 12 May 2017, a self-spreading ransomware worm moved between Windows computers around the world, spreading to more than 150 countries. It encrypted files and demanded a ransom of about 300 US dollars (BBC).

Impact on hospitals. According to the npj Digital Medicine study, 34 hospital trusts in England were infected and 46 more were affected through precautionary action or shared systems. Infected trusts were locked out of digital systems and medical devices such as MRI scanners, reverted to manual processes for blood results and notes, and five acute trusts diverted emergency ambulances. Infected hospitals saw about 6% fewer admissions per day, and 13,500 outpatient appointments were cancelled. The study valued lost activity at £5.9 million. NHS England identified 6,912 appointments cancelled as a direct result and estimated about 19,000 affected in total, including at least 139 urgent referrals for patients who potentially had cancer. The study found no difference in mortality, which its authors call a crude measure of harm.

Why it spread. Microsoft explained that WannaCry used publicly available exploit code for the SMBv1 vulnerability fixed in MS17-010 on 14 March 2017. The npj study cites an official UK review which concluded that none of the affected NHS organisations had applied that update, despite advice to do so, and the BBC reported that firewalls could also have been better managed. The attack was stopped within 24 hours, after a security researcher registered a domain name that acted as a kill switch for the worm's spreading.

Lesson. Patch known critical flaws fast, isolate what cannot be patched, and test the incident plan. The BBC's analysis noted that a central plan existed but had not been properly communicated or tested in the trusts. Our WannaCry retrospective covers the technical side in more depth.

Change Healthcare, February 2024

What happened. Change Healthcare, part of UnitedHealth Group, handles about 15 billion health care transactions a year and is involved in roughly one in three US patient records, according to the American Hospital Association. On 12 February 2024, attackers used compromised credentials on a remote access portal that lacked multi-factor authentication. On 21 February they deployed ransomware (Healthcare Dive).

Impact on providers and claims. The attack disrupted claims processing, payments to providers, prior authorisation requests and eligibility checks. In an AHA survey of nearly 1,000 hospitals run from 9 to 12 March 2024, 94% reported a financial impact (more than half called it significant or serious), 83% reported an impact on cash flow and 74% reported an impact on direct patient care. Nearly 40% said patients had difficulty getting care because of delays in prior authorisation. Switching clearinghouses was difficult or very difficult for 67%, and 81% found workarounds only somewhat successful.

Bar chart of AHA survey results: 94 percent of hospitals reported financial impact, 83 percent cash flow impact, 74 percent impact on direct patient care, 67 percent found switching clearinghouses difficult, and nearly 40 percent saw patients struggle to access care

Figure 7. How one supplier's outage reached US hospitals, from the AHA survey of March 2024.

Recovery. By 7 March, UnitedHealth Group reported that electronic prescribing was fully functional with pharmacy claims and payment transmission available, and planned to begin testing connectivity to its medical claims network from 18 March. It advanced more than 9 billion US dollars to providers during 2024 to cover the payment gap (CNBC). The company's costs from the attack reached 3.1 billion dollars in 2024, against an initial estimate of 1.6 billion (Healthcare Dive). Its chief executive confirmed a ransom of 22 million dollars had been paid. The breach affected about 192.7 million people, the largest health data breach reported in the US (TechTarget).

Lesson. Multi-factor authentication on every external login, with no exceptions for acquired companies or legacy portals. And for providers: map which suppliers carry your revenue and prescriptions, and plan for losing them. Our earlier Change Healthcare analysis goes deeper.

Synnovis and London hospitals, June 2024

What happened. On 3 June 2024, Synnovis suffered a ransomware attack that affected almost all of its IT systems and disrupted many pathology services. Data was taken "in haste and in a random manner" from working drives; Synnovis says no data was taken from its primary lab databases. On 20 June the attackers published the stolen data, and Synnovis obtained an injunction against further publication (Synnovis).

Impact on hospitals and patients. More than 10,000 appointments were cancelled at King's College Hospital and Guy's and St Thomas' over the following months, the two trusts worst affected, and a significant number of GP practices in London could not order blood tests for their patients. By August, more than 60 laboratory IT systems had been rebuilt or were being rebuilt, and full restoration of blood transfusion services was planned for early autumn (PA Media). In July, NHS Blood and Transplant issued an amber alert, only the second ever, after national stocks of O negative and O positive blood fell to what it called "unprecedentedly low levels".

In June 2025, King's College Hospital confirmed that a patient had died unexpectedly during the incident, and that its safety investigation found a number of contributing factors, including "a long wait for a blood test result due to the cyber-attack impacting pathology services". The Health Service Journal reported nearly 600 incidents linked to the attack, with patient care affected in 170, one case of severe harm and 14 of moderate harm (BBC).

Recovery. By late autumn 2024, access to all services that were available before the attack was restored, and none of the affected infrastructure remains. The forensic investigation of the stolen data took more than a year, ending in summer 2025, and the organisations whose data was affected were notified by the end of November 2025. Under UK data protection law, those organisations, as data controllers, decide whether to notify patients (Synnovis).

Lesson. A pathology supplier is part of the clinical system. Its security, its recovery time and the hospital's manual fallback for losing it need the same attention as the hospital's own systems.

What keeps patients safe when systems go down?

Downtime procedures keep patients safe: read-only copies of key records, paper forms for orders and results, runners between departments, clear rules for diverting ambulances, and a plan to re-enter paper records afterwards. Hospitals that drill these recover faster, as Romania's 2024 hospital outage showed.

Hospitals already plan for power cuts and planned system upgrades. A cyberattack is harder for three reasons: it can last weeks, it can take out the downtime tools themselves, and nobody knows at first how far it has spread.

Timeline of a hospital downtime from the first hour to recovery: switch to read-only records and paper, set up runners and manual lab processes, decide on diversion, then restore systems and re-enter paper records

Figure 8. A downtime from first hour to recovery. Each phase needs an owner and a drill.

What a good downtime plan contains:

  • Read-only access to critical data. Standalone downtime computers, kept off the main network and updated regularly, with current patient lists, medication records, allergies and recent results. Plus printed census and medication sheets on a schedule.
  • Paper forms for everything. Orders, results, medication administration, blood requests, admissions and discharges, stocked on every ward and in a central store.
  • Manual lab and blood bank processes. Paper request forms, phoned or hand-carried critical results, manual blood matching procedures and a plan with the blood service for emergency stock.
  • Communications that do not depend on the network. Printed phone lists, radios or pagers, and a command centre.
  • Diversion and transfer rules. Clear criteria for when to divert ambulances or transfer patients, agreed in advance with neighbouring hospitals and ambulance services.
  • Recovery and re-entry. Who enters paper records into the restored system, in what order, and how results are reconciled.

The Romanian case. In February 2024, after ransomware spread through hospitals using a shared hospital information system, more than 100 hospitals disconnected from the internet on instruction from the country's cyber security incident team. Doctors created offline patient registers, asked labs for paper results and used spreadsheets. The attackers asked for 160,000 euros, and a decision was taken not to pay. Most hospitals had recent backups. Within five days most were back online and operating close to normal, with no reported deaths or serious harm. It still took weeks to enter the paper records, and some data was lost for good (BBC).

The neighbours. A hospital's outage spills over. A 2023 study in JAMA Network Open looked at two emergency departments in San Diego during a month-long ransomware attack on a separate nearby health system in May 2021. The unaffected departments saw significant increases in daily census, ambulance arrivals, waiting times, patients leaving without being seen, and stroke code activations (102 against 59 in the prior four weeks). The authors concluded that such attacks should be treated as a regional disaster. Regional plans need to include the hospital next door.

How can a hospital detect an attack early?

Hospitals detect attacks early by knowing every asset on the network, collecting logs from identity, remote access, servers and clinical systems in one place, and watching for the specific signs that precede ransomware: new privileged accounts, remote sessions to domain controllers, backup deletion and large outbound transfers.

The practical order of work:

  1. Asset inventory first. You cannot monitor what you do not know exists. Combine IT asset management, clinical engineering records and network discovery into one inventory with an owner for each system.
  2. Centralise the right logs. Identity provider and domain controller logs, VPN and remote access logs, email security, endpoint detection on servers and workstations, firewall logs between zones, EHR access logs and backup system logs.
  3. Watch the pre-ransomware signals. Use the table below.
  4. Monitor the clinical network passively. Network detection that understands HL7 and DICOM can spot new connections to modalities or interface ports without touching fragile devices.
  5. Test the alerts. Run tabletop exercises and authorised purple team tests to confirm the alerts fire and someone answers them at 3 a.m. on a Sunday.
SignalWhat it can mean
Login to remote access from a new country or deviceStolen credentials
Vendor account active outside its maintenance windowCompromised vendor access
New member of a domain admin groupPrivilege escalation
Remote desktop session to a domain controller from a workstationHands-on-keyboard attacker
New remote support tool installed on a serverAttacker persistence
Backup jobs failing or snapshots deletedPreparation for encryption
Large outbound transfer from a file or database serverData theft before extortion
Burst of SMB connections from one workstationWorm or scanning
One account opening thousands of patient chartsInsider misuse or account takeover

What do HIPAA and GDPR mean for a hospital breach?

In plain language, HIPAA requires US healthcare providers to protect electronic health information with administrative and technical safeguards and to notify affected people within 60 days of discovering a breach. GDPR treats health data as a special category and requires EU and UK organisations to notify regulators within 72 hours.

These are short, plain summaries, not legal advice. Our guide to HIPAA and GDPR differences goes further.

HIPAA in plain language

HIPAA applies to US covered entities (healthcare providers, health plans and clearinghouses) and their business associates, the vendors that handle health data for them.

  • Safeguards. The Security Rule requires administrative safeguards, including risk analysis and risk management, security incident procedures and contingency plans with data backup, disaster recovery and emergency mode operation (45 CFR 164.308). It also requires technical safeguards: access control, audit controls, integrity protection, person or entity authentication and transmission security (45 CFR 164.312). A downtime plan is a HIPAA requirement, not an extra.
  • What counts as a breach. Any access, use or disclosure of protected health information that the rules do not permit is presumed to be a breach, unless the organisation can show a low probability that the information was compromised, based on a documented risk assessment (45 CFR 164.402).
  • Ransomware. The US health privacy regulator has said that a ransomware attack that encrypts patients' electronic health information is presumed to be a HIPAA breach, as reported by HIPAA Journal.
  • Deadline. Affected individuals must be notified without unreasonable delay and in no case later than 60 calendar days after discovery (45 CFR 164.404).

GDPR in plain language

GDPR applies to organisations processing personal data of people in the EU. In the UK, UK data protection law applies; the Synnovis case below shows how it assigns responsibility.

  • Health data is special. "Data concerning health" is defined in Article 4 and listed as a special category in Article 9, which may only be processed under specific conditions such as the provision of health care.
  • Security. Article 32 requires security appropriate to the risk, including the ability to restore availability and access to personal data in a timely manner after an incident, and regular testing of security measures. Backups and recovery are a GDPR duty.
  • 72 hours. A controller must notify its supervisory authority within 72 hours of becoming aware of a breach where feasible, unless it is unlikely to result in a risk to people (Article 33).
  • Telling patients. If the breach is likely to result in a high risk to people, the controller must also tell them without undue delay (Article 34).
  • Controllers and processors. In the Synnovis case, the lab provider was the processor, so each hospital and practice, as controller, was responsible for deciding whether and how to notify its patients.

In the EU, the NIS 2 Directive adds cybersecurity risk management and incident reporting duties for organisations in sectors including health (overview). Its Article 21 lists the measures expected, including incident handling, business continuity with backup management and crisis management, supply chain security covering direct suppliers and service providers, and multi-factor authentication where appropriate. That list reads like a summary of this guide.

Where does the healthcare software supply chain break?

The healthcare software supply chain breaks where hospitals cannot see inside the software they buy: open source libraries in EHRs and portals, Windows images inside PACS and lab systems, embedded components in devices, and hosted services run by suppliers. SBOMs and supplier security evidence turn that blind spot into an inventory.

Layered diagram of the healthcare software supply chain, from open source libraries and operating systems through vendor products and hosted services to the hospital, showing where SBOMs and supplier evidence apply

Figure 9. Where the supply chain sits inside a hospital, and where SBOMs and supplier evidence help.

The chain has more links than most people expect:

  • Open source inside commercial products. An EHR, a PACS viewer or a patient portal carries hundreds of open source libraries. When a widely used library gets a critical vulnerability, the hospital needs to know within hours which products contain it, and the honest answer without an SBOM is usually "we have to ask every vendor".
  • Operating systems inside appliances. Many clinical systems ship as a server or virtual machine image with an operating system the vendor chose and patches on its own schedule.
  • Shared hospital software. The Romanian case shows that one hospital information system used by many hospitals is a single point of compromise for all of them.
  • Hosted and outsourced services. Clearinghouses, pathology and hosted EHRs, where the hospital's security depends entirely on the supplier's.
  • Updates. Software updates are a trusted path into every customer. They need signing, provenance and verification on both sides.

What an SBOM gives a hospital. A software bill of materials lists the components in a product with their versions, usually in the CycloneDX or SPDX formats. With SBOMs for its major systems, a hospital can match a new vulnerability against its whole estate in minutes, prioritise by which systems are internet-facing or clinical, and ask the vendor a precise question instead of a vague one. A VEX (Vulnerability Exploitability eXchange) statement from the vendor then says which listed vulnerabilities actually affect the product, which keeps the list honest.

What vendors are expected to do. IEC 81001-5-1:2021 defines the secure life cycle requirements for developing and maintaining health software, aligned with IEC 62443-4-1 (IEC). That includes managing third-party components, which in practice means an SBOM and a vulnerability handling process. Medical device makers face further rules, covered in our medical device SBOM guide. Hospitals increasingly put SBOM delivery, vulnerability response times and security evidence into contracts.

Hardening checklists for hospitals, clinics and health IT vendors

Hardening a hospital follows a clear order: multi-factor authentication and patching at the edge, an asset inventory, network segmentation into zones and conduits, controlled vendor access, monitoring, tested offline backups and a drilled downtime plan. Vendors add secure development, SBOMs and fast, validated fixes.

Defence layers diagram with six rings around patient care: identity and access, edge and patching, segmentation, monitoring, backup and recovery, and downtime procedures, with supplier assurance across all of them

Figure 10. Defence layers for a hospital. No single layer is enough on its own.

Segmenting a hospital: the Purdue model and IEC 62443 zones and conduits

Industrial control systems have long used two ideas that fit hospitals well. The Purdue model is a reference architecture from the 1990s that arranges a network into levels, from physical equipment at the bottom through control and operations up to enterprise IT, with a buffer zone (a DMZ, demilitarised zone) between the operational levels and the business network (Fortinet). IEC 62443, a series of standards for securing automation and control systems (ISA), groups assets with similar security needs into zones and controls every communication path between them as a conduit (Fortinet). ISA notes that the series applies to all automation and control systems, not only industrial ones.

Applied to a hospital, a workable layout is:

LevelHospital zoneExamplesAllowed conduits
5Internet and partnersPatient portal users, payers, partnersOnly to the DMZ
4Enterprise ITEmail, finance, HR, office workstationsTo DMZ services; to clinical apps only through defined application ports
DMZBrokered servicesRemote access gateway, portal web tier, FHIR gateway, file transfer, update serversInbound from levels 4 and 5; outbound to named clinical services only
3Clinical applicationsEHR servers, interface engine, PACS archive, LIS, pharmacy serversTo devices on defined protocols; to DMZ services
2Clinical workstations and central stationsNursing stations, radiology reading, monitoring centralsTo level 3 applications only
1 and 0Clinical devicesModalities, analysers, pumps, monitors, dispensing cabinetsOnly to their own central station, archive or LIS; maker updates through the DMZ

Building systems such as HVAC, lifts, pneumatic tubes and medical gas alarms get their own zone with the same rules. Vendor access arrives through the DMZ gateway and is allowed only to the zone the vendor supports.

Hospital network zones arranged by Purdue level with conduits between them: internet, DMZ with remote access gateway, enterprise IT, clinical applications, clinical workstations, and device zones for imaging, lab, infusion and building systems

Figure 11. A hospital network arranged into IEC 62443 style zones and conduits, by Purdue level.

Hospitals

Identity and access

  • Multi-factor authentication on every remote access path, email, cloud console and privileged account, with no exceptions for legacy portals or acquired businesses.
  • Phishing-resistant authentication (security keys or passkeys) for administrators.
  • Separate admin accounts, tiered administration and protected domain controllers.
  • Fast badge-based login for shared clinical workstations.

Edge and patching

  • A complete list of internet-facing systems, scanned continuously, with no PACS, DICOM, RDP or SMB exposed.
  • Emergency patching for remote access gateways, file transfer tools and anything on known exploited vulnerability lists.
  • Supported operating systems as a purchasing requirement.

Inventory and segmentation

  • One inventory across IT, clinical engineering and facilities, with owners.
  • Zones and conduits as in the table above, starting with the most critical clinical applications and the oldest devices.
  • SMBv1 and RDP blocked between zones.

Vendor and supplier access

  • One managed gateway, named accounts, multi-factor authentication, session approval and recording, no standing access.
  • Security terms in every contract: notification times, SBOMs, vulnerability handling, recovery times, audit rights.
  • A dependency map of critical suppliers, with a downtime plan for losing each one.

Monitoring and response

  • Central logging of identity, remote access, servers, firewalls between zones, EHR access and backups.
  • Alerts for the pre-ransomware signals in the detection table, answered around the clock.
  • An incident response plan with named clinical, technical, legal and communications leads, and a retainer with an incident response firm.

Backup and recovery

  • Offline or immutable backups of the EHR, PACS, LIS, directory and key configurations, kept separate from the main domain.
  • Regular full restore tests, timed, including rebuilding the directory from scratch.
  • Recovery priorities agreed with clinicians in advance.

Downtime

  • Downtime computers, printed reports, paper forms and manual lab and blood bank procedures.
  • A drill at least yearly that runs for a full shift, including re-entry of paper records.
  • Regional plans with neighbouring hospitals and ambulance services.

Clinics and small practices

  • Turn on multi-factor authentication for email, the practice management system and any remote access. This single step blocks most account takeovers.
  • Use a hosted EHR from a vendor that can show you a SOC 2 report and a recovery plan, and keep a printed schedule and medication list for each day.
  • Keep computers on supported operating systems with automatic updates, and replace anything that cannot be updated.
  • Back up to a location that a compromised computer cannot delete, and test a restore every quarter.
  • Never expose remote desktop or an imaging server to the internet. Use the vendor's secure remote access instead.
  • Know whom to call: your IT provider, your EHR vendor's security contact and your cyber insurer.

Health IT vendors

  • Follow a secure development life cycle such as IEC 81001-5-1, with threat modelling, code review and security testing in every release.
  • Scan your own code (SAST), your dependencies (SCA) and your repositories and images for secrets on every build.
  • Produce an SBOM for every release in CycloneDX or SPDX, and publish VEX statements so customers know which vulnerabilities affect them.
  • Sign releases and record build provenance, so customers can verify updates.
  • Support current operating systems and give customers a tested patch path within agreed times.
  • Remove default credentials from products, and ship with secure defaults: TLS on, multi-factor authentication available, logging on.
  • For hosted services, enforce multi-factor authentication on every external login, segment customer environments, and rehearse your own restore. Your outage is your customers' downtime.
  • Run a coordinated vulnerability disclosure programme and answer researchers quickly, as OpenEMR did.

Patients and families

  • Use the hospital's patient portal with a unique password and turn on two-step verification if it is offered.
  • Be cautious of calls or emails after a publicised breach that ask for payment details or passwords. Hospitals will not ask for those.
  • If your appointment is postponed during an incident, keep the new date and ask your clinician how to get urgent care in the meantime.

How Safeguard helps hospitals and health IT vendors

Safeguard covers the software supply chain behind a hospital in one platform: SBOMs for the software you build and buy, scanning of vendor binaries and virtual appliance images, reachability-ranked vulnerabilities, SCA, SAST and secrets scanning for code you own, fix pull requests and patched packages, supplier risk and HIPAA and GDPR evidence, as SaaS, in your VPC or on premises.

Flow diagram of Safeguard for health IT: inputs from source repositories, vendor installers, virtual appliance images and containers flow into SBOMs and vulnerability analysis ranked by reachability, then fix pull requests gated by CI, Gold patched packages, supplier risk scoring and compliance evidence

Figure 12. One platform from inventory to fix and evidence, for hospitals and the vendors that serve them.

The hospital problems above are mostly about network and identity. The supply chain is where Safeguard fits. Health IT teams often stitch together a source code scanner, an SBOM manager, a vulnerability tracker, a third-party risk tool and a GRC (governance, risk and compliance) platform, 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 asks for.

See inside the software you buy

In the Safeguard console, Integrations > Upload App Binary accepts Windows EXE, MSI and DLL files, Android APKs, iOS IPAs, virtual machine 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 (exploit prediction) and known exploited enrichment, records licences, and creates a project with a full SBOM.

For a hospital, that covers the vendor installers for clinical workstation software, the virtual appliance images many PACS, interface engine and lab vendors ship, and containerised services. You can see what is inside a product before you deploy it, and check your whole estate when a library vulnerability is announced. Raw device firmware may need to be extracted into its component files first; Safeguard does not claim to parse every raw firmware format directly. Heavy obfuscation or encryption 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 health IT vendors, and for hospitals that build their own portals, integrations and apps, Safeguard runs software composition analysis (SCA) across the full transitive dependency tree, static application security testing (SAST) on your own code for flaws such as injection and missing authorisation, and secrets scanning that catches API keys, FHIR client secrets and cloud credentials before they ship. Dynamic application security testing (DAST) tests patient portals and FHIR APIs on verified targets you own. Package Firewall sits at install time and can block malicious, typosquatted and dependency confusion packages before they reach a build.

Rank by reachability, not by count

An EHR or portal can carry hundreds of known vulnerabilities in its libraries. Reachability analysis shows which of them your code can actually reach, so a flaw in an internet-facing request parser rises above one in an unused function. For a team planning the next validated release, that turns a long list into a short one you can act on.

SBOMs and evidence customers can use

Every project produces an SBOM you can export in CycloneDX or SPDX, among many other formats. Attestation and signing with SLSA, Sigstore and in-toto record build provenance. CSAF and VEX documents can be uploaded and linked, so the SBOM a vendor gives a hospital comes with a clear statement of which vulnerabilities actually affect the product.

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. For a vendor shipping into hospitals, the fix enters your normal review and release process as a tracked change.

For open source components where upstream has not shipped a fix, or the fix is in a version you cannot move to on a supported release 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. A minimal backported fix is far easier for a hospital to accept and validate than a major version upgrade.

When the vulnerable code sits inside a product you buy rather than build, Safeguard shows exactly which products and versions contain it, so you can press the vendor with evidence and put compensating controls, such as tighter zone rules, in place while you wait.

Supplier risk and compliance evidence

Safeguard's Third-Party Risk Manager lets a hospital request SBOMs from EHR, PACS, lab and device vendors, 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 risk analyses, audits and the security questionnaires hospitals send vendors. Guard adds runtime guardrails on Linux servers, inspecting system calls, library calls and network egress.

Deploy where your data lives

Health data and source code are sensitive. Safeguard runs as SaaS in US and EU regions, in your own VPC or fully on premises. It 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

Hospitals often use clinical asset visibility platforms such as Claroty, Armis or Ordr to discover and monitor devices on the network, and those products are the right tool for that job. Vendors often combine a source SCA tool such as Snyk or Black Duck with a separate SBOM manager and a GRC platform such as Vanta or Drata. Safeguard's case is consolidation and remediation across the software supply chain: one inventory across source, binaries, appliance images and containers; reachability-ranked vulnerabilities; fix pull requests and patched packages instead of tickets; and supplier risk and compliance evidence from the same data. It sits well alongside a network visibility platform, which sees the devices, while Safeguard sees and fixes the software inside them.

The takeaway

Most hospital attacks are not sophisticated. They use a password without a second factor, a patch that was available for months, a database answering anyone on the internet, or a supplier that nobody planned to lose. WannaCry, Change Healthcare and Synnovis each began with one of those, and each cost patients time and care.

The defences are known and they work in order. Multi-factor authentication and patching at the edge. An inventory everyone shares. Zones and conduits that stop one infected laptop from reaching every scanner and lab system. Vendor access that is controlled and recorded. Monitoring that notices the days between entry and encryption. Backups that the attacker cannot reach. And a downtime plan that clinicians have actually practised.

Vendors carry the other half. Hospitals can only defend software they can see into, so an SBOM, a VEX statement and a fast, validated fix are now part of what a health IT product is.

Try it yourself

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

Test yourself

How do hospitals mostly get hacked?

Mostly through ordinary IT weaknesses: stolen or phished credentials on remote access without multi-factor authentication, unpatched internet-facing systems, old Windows, flat clinical networks, and suppliers whose compromise spreads into the hospital.

Why are PACS and DICOM imaging systems exposed?

Many imaging archives are reachable from the internet with no authentication, so anyone can query them for patient names and scans. Researchers found thousands of such servers in 2019, 2023 and again in late 2025.

How did Change Healthcare begin?

It began with a remote access portal without multi-factor authentication. It showed how one supplier's missing control can freeze payments and prescriptions across a country.

What did WannaCry and Synnovis each teach hospitals?

WannaCry showed how unpatched Windows on flat networks turns a worm into a national outage. Synnovis showed how a pathology supplier's outage harms patients for months.

What are the usual stages of hospital ransomware, and where is the defender's best chance?

Initial access, days of quiet movement and privilege escalation, theft of data, then encryption of servers and backups. The days between entry and encryption are the best chance to stop it.

What does a good downtime plan keep patients safe with?

Read-only copies of key records, paper forms for orders and results, runners between departments, clear rules for diverting ambulances, and a plan to re-enter paper records afterwards.

Frequently asked questions

How do most hospital cyberattacks start?

Most start with a login, not an exploit: a phished or reused password on remote access or email without multi-factor authentication. The rest mostly start with an unpatched internet-facing system, such as a remote access gateway or file transfer tool. Change Healthcare's 2024 attack began with compromised credentials on a portal that lacked multi-factor authentication.

Why do hospitals still run old versions of Windows?

Clinical systems and devices are validated as a whole by their makers on a specific operating system, so hospitals cannot upgrade them without the maker's support. Some older products never get one. Unit 42 found 83% of imaging devices it observed in 2020 ran unsupported systems. Hospitals compensate by isolating them in tightly controlled network zones.

What happens to patients when a hospital's computer systems go down?

Staff switch to downtime procedures: paper orders and results, printed medication lists, runners and phoned results. Planned appointments and operations may be postponed, and ambulances may be diverted. During WannaCry, five English acute trusts diverted ambulances, and infected hospitals cancelled about 13,500 outpatient appointments in a week.

Can a cyberattack on a hospital harm patients?

Yes. After the 2024 Synnovis pathology attack, King's College Hospital confirmed that a long wait for a blood test result caused by the attack was one of several factors in a patient's death. Reporting also counted one severe and 14 moderate harm cases. Delays to tests, surgery and emergency care are the main mechanism.

What was the Change Healthcare cyberattack and why did it affect so many providers?

In February 2024, ransomware hit Change Healthcare, a US clearinghouse that handles about 15 billion transactions a year. Claims, payments, eligibility checks and prior authorisations stalled. In an AHA survey, 94% of hospitals reported financial impact and 74% an impact on patient care. About 192.7 million people's data was affected.

Is a ransomware attack a HIPAA breach?

Usually, yes. HIPAA presumes any unpermitted access to protected health information is a breach unless a documented risk assessment shows a low probability of compromise, and the regulator has said ransomware encryption of health data is presumed a breach. Affected people must be told within 60 days of discovery.

What does GDPR require after a hospital data breach?

Health data is a special category under GDPR. The controller must notify its supervisory authority within 72 hours of becoming aware of a breach where feasible, unless it is unlikely to create risk, and must tell affected patients without undue delay if the risk to them is high. Article 32 also requires timely restoration of data.

Should hospitals ask their software vendors for an SBOM?

Yes. A software bill of materials lists every component in a product, so when a library vulnerability is announced the hospital can find affected systems in minutes rather than polling every vendor. Ask for SBOMs in CycloneDX or SPDX at purchase and each release, with VEX statements and agreed patch times in the contract.

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.