WannaCry showed hospitals two patch gaps: their own unpatched PCs and devices only vendors could update
In the week after WannaCry began on May 12, 2017, the fix for supported Windows was a March patch. In our reading, hospitals faced two different gaps: slow patching of computers they controlled, and medical equipment whose software only its manufacturer could change.
Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.
Two-minute summaryThe dek and the first sentence of each section, quoted from the article
In the week after WannaCry began on May 12, 2017, the fix for supported Windows was a March patch. In our reading, hospitals faced two different gaps: slow patching of computers they controlled, and medical equipment whose software only its manufacturer could change.
- The week The week of May 15, 2017, was the first full working week after WannaCry.
- The hospital’s own machines: a patch that was available The first part asks why computers a hospital controlled were exposed when a fix existed.
- The manufacturer’s machines: equipment the hospital could not patch The second part asks why some machines stayed exposed whatever the hospital did.
- The network between them: containment in place of patching The third part asks how hospitals protected machines they could not patch in time.
- The clinical work on top: keeping care running The fourth part asks what kept patients safe while machines were locked or isolated.
- How the week reached different health organizations The four parts applied across health care.
- Objections, in order The NAO found the opposite for the NHS.
- What hospital technology and security leaders could check The four parts translate into checks a hospital can run today, for SMB or any flaw that reaches both its computers and its devices.
- With hindsight The NAO published its investigation in October 2017. [5] Its account, used throughout this article, is a detailed official public record of a health system under WannaCry.
The week
The week of May 15, 2017, was the first full working week after WannaCry. US-CERT’s alert, first released on Friday, May 12, and revised on May 16, set out what was known. [1] The ransomware variant, known as WannaCry, WCry or Wanna Decryptor, was discovered on the morning of May 12, with initial reports beginning around 4:00 AM EDT. [1] It had spread rapidly over several hours. [1]
The alert cited open-source reports of tens of thousands of infections in over 150 countries, including the United States, the United Kingdom, Spain, Russia, Taiwan, France and Japan. [1] The software could run in as many as 27 languages. [1] Open-source reporting put the ransom demand at .1781 bitcoins, roughly $300. [1] The alert was the product of work between DHS’s National Cybersecurity and Communications Integration Center and the FBI. [1]
The route in was a known flaw. Initial reports indicated the attackers were reaching enterprise servers by exploiting a critical Windows SMB vulnerability. [1] Microsoft had released a security update for that flaw, MS17-010, on March 14, 2017. [1] [2] Microsoft’s bulletin said the most severe of the vulnerabilities could allow remote code execution if an attacker sent specially crafted messages to a Microsoft Server Message Block 1.0 server. [2] It rated the update Critical for all supported releases of Windows. [2]
Supported was the important word. On May 13, 2017, Microsoft released patches for Windows XP, Windows 8 and Windows Server 2003, according to the alert. [1] Those releases followed the outbreak that began on May 12. [1] Microsoft’s March bulletin had described its rating in terms of supported releases. [2]
US-CERT’s analysis described the malware. Three files submitted for analysis were confirmed as components of the campaign. [1] The first was a dropper that ran the ransomware and propagated through the MS17-010 exploit, which the alert also called EternalBlue. [1] It scanned local IP ranges and tried to connect on UDP ports 137 and 138 and TCP ports 139 and 445. [1] The other two files held the components that encrypted the victim’s files. [1]
The alert explained why the spread slowed. The dropper tried to connect to a hard-coded web address. If the connection succeeded, it stopped. [1] Researchers learned that registering the domain and redirecting it to a monitored server caused the malware to terminate. [1] For that reason, US-CERT recommended that administrators not block traffic to that domain. [1]
The recommendations started with the patch. Apply the MS17-010 update dated March 14, 2017. [1] If the patch could not be applied, consider blocking all versions of SMB at the network boundary. [1] That meant blocking TCP port 445, with UDP ports 137 and 138 and TCP port 139, on all boundary devices. [1] The alert warned that disabling or blocking SMB may obstruct access to shared files, data or devices, and that benefits should be weighed against disruption. [1]
The rest of the list was general hygiene. It covered spam filtering and email authentication, scanning email for threats, automatic antivirus scans, least privilege for accounts and file shares, disabling Office macros sent by email, staff education and penetration testing at least yearly. [1] The alert also said paying a ransom does not guarantee decryption and does not remove the infection. [1]
For hospitals, one more document framed the week. In December 2016, FDA had issued final guidance on postmarket cybersecurity for medical devices. [3] It said the majority of manufacturer actions to address cybersecurity vulnerabilities are what it called routine updates and patches. [3] Those are generally treated as device enhancements for which FDA does not require advance notification or reporting under 21 CFR part 806. [3]
A hospital runs two kinds of computers. Some belong to its IT department: desktops, servers, the machines that run scheduling and records. Others belong, in practice, to the companies that made them: imaging systems, laboratory analyzers, monitors and other equipment with an operating system inside. WannaCry used one flaw against both. [1] The fix for that flaw had been available for supported Windows since March. [2] The path to applying it was different for each kind of machine.
Our argument is that WannaCry exposed two separate patch gaps in hospitals, and that the two needed different remedies. The first gap was the hospital’s own speed in patching computers it controlled. The second was equipment whose software only its manufacturer could change. Treating them as one problem, under one patch metric, hid the second. The network between the two kinds of machine, and the clinical work that depends on both, decided how much harm either gap could do.
We take the week in four parts:
- The hospital’s own machines: why supported, patchable computers were still exposed.
- The manufacturer’s machines: why some equipment could not be patched by the hospital at all.
- The network between them: how containment substituted for patching, and what it cost.
- The clinical work on top: what kept care running when machines were locked or isolated.
The first two parts are about the two gaps. The third is about the control that covers both. The fourth is about consequences. Together they run from the flaw to the patient. After the four parts, we look at how the event reached different kinds of health organization, answer objections in order, and list what hospital technology and security leaders could check.
A note on sources and our reading. We describe the attack from US-CERT’s alert as it stood on May 17, 2017, and the flaw from Microsoft’s bulletin. [1] [2] We use the UK National Audit Office’s later investigation of the NHS as a detailed official account of how a health system experienced the attack. [5] US hospitals and the NHS differ in structure, and what we say about US organizations is our reading, not a finding in that report.
The hospital’s own machines: a patch that was available
The first part asks why computers a hospital controlled were exposed when a fix existed. Microsoft rated MS17-010 Critical for all supported releases of Windows and released it on March 14, 2017. [2] US-CERT’s first recommendation was to apply it. [1]
What the NHS evidence showed
The UK National Audit Office’s investigation, which covered the NHS in England, is direct on this point. NHS Digital told the NAO that the majority of NHS devices infected were unpatched but on the supported Windows 7 operating system. [5] Unsupported Windows XP devices were in the minority of identified issues. [5] Trusts on Windows 7 could have protected themselves by applying the March patch, and NHS Digital had issued alerts on March 17 and April 28 asking trusts to apply it. [5]
That finding cuts against the familiar story about WannaCry and old software. Old software mattered. The larger exposure, in that health system, was current software left unpatched. [5]
Why a patch can sit for weeks
A hospital can have reasons to move slowly on a patch. A patch can break a clinical application. Testing it against every application takes time and staff. Some systems cannot be restarted during clinical hours, and a hospital never closes. In our reading, these reasons are real, and they explain a lag. They do not explain a lag long enough to miss two national alerts for a critical flaw. [5]
In our reading, the gap lies in prioritization. A hospital that patches every month at the same pace, regardless of severity, treats a critical remote code execution flaw like any other update. A hospital with a fast lane for critical flaws on exposed services can move faster without testing everything faster.
What a fast lane needs
A fast lane needs four things. A trigger, such as a vendor’s Critical rating or a national alert. [2] [5] A list of machines running the affected service. A reduced test, focused on whether core clinical applications still start. And authority to restart machines outside the normal window. Each element is a decision made in advance. Without them, the response to a critical alert starts with a meeting.
Which machines ran the vulnerable service
The flaw sat in the SMBv1 server, and Microsoft’s update corrected how SMBv1 handles specially crafted requests. [2] A machine exposed to the worm was one that accepted SMBv1 connections. In our reading, the first inventory question was therefore narrower than which machines ran Windows. It was which machines ran the SMBv1 server and could be reached on port 445. [1]
That narrower question had a second answer besides patching. Where a machine did not need to share files over SMBv1, the service could be turned off. Where it did, the machine moved to the top of the patch queue. A hospital that knew which machines served files to which others could make both choices quickly. A hospital that did not had to treat every Windows machine alike.
Assessment without enforcement
The NAO’s account shows a second weakness in the first part. Before the attack, NHS Digital offered a voluntary on-site cybersecurity inspection to hospitals. [5] By May 12, it had inspected 88 of 236 trusts, and none had passed. [5] NHS Digital could not mandate a local body to take remedial action, even if it had concerns about that organization’s vulnerability. [5]
NHS Digital’s own review, as the NAO reported it, concluded that its advice was mostly followed by organizations with relatively mature security arrangements, while vulnerable trusts were not acting to improve. [5] It also found that trusts in general had not identified cyber security as a risk to patient outcomes and had overestimated their readiness. [5] In our reading, that is the pattern any advisory regime risks. The organizations that read alerts carefully are the ones that need them least.
The weekend before the Monday
The NAO recorded that national and local NHS staff worked overtime over the weekend of May 13 and 14 to resolve problems and to prevent a fresh wave of infections on Monday, May 15. [5] On May 12, NHS England’s emergency teams asked providers for a positive return by midnight confirming, among other things, that those not attacked had implemented the patch. [5]
That is a fast lane built in an emergency. In our reading, any hospital is better served by building it in advance. A Monday morning wave is a predictable risk after a Friday outbreak, because staff return and switch on machines that were off over the weekend.
Measuring the right thing
A hospital’s patch report can show a high overall compliance rate and still hide the exposure that matters. The NAO reported that by the evening of Sunday, May 14, 3,486 GP practices, 44 percent, had applied the necessary patches. [5] That is a measure of one flaw, across one population, over a weekend. In our reading, it is the kind of measure a hospital should be able to produce for any critical flaw: how many exposed machines are patched, by when.
The manufacturer’s machines: equipment the hospital could not patch
The second part asks why some machines stayed exposed whatever the hospital did. The answer lies in who controls the software inside medical equipment.
What the NHS evidence showed about devices
The NAO found that some trusts had problems with medical equipment, such as MRI scanners, that had Windows XP embedded in them. [5] It said this equipment is generally managed by the system vendors and local trusts are not capable of applying updates themselves. [5] Support from the vendors of these devices was often poor, according to NHS England and NHS Digital. [5]
The scale was not trivial. As of May 19, 2017, NHS England had identified 1,220 pieces of diagnostic equipment that had been infected, 1 percent of all such NHS equipment. [5] The figure did not include devices disconnected from IT systems to prevent infection. [5] Trusts told the NAO of disruption such as being unable to send MRI scan results to clinicians in other parts of the hospital. [5]
The manufacturers’ notices
In the United States, ICS-CERT’s WannaCry alert collected notices from manufacturers whose products use Windows. [4] Its list appears in an update that followed one published May 31, 2017. [4] It included makers of medical and laboratory equipment such as Becton, Dickinson and Company, Beckman Coulter, Dräger, Johnson & Johnson, Medtronic, Philips, Smiths Medical, Spacelabs Healthcare and Toshiba Medical Systems. [4] Each had proactively issued customer notifications with recommendations. [4]
In our reading, those notices were the only authoritative source of whether a given device was affected and what to do. A hospital could not inspect the software inside an imaging system to find out. It needed the maker’s statement, model by model.
Why the hospital could not act alone
The barrier was not FDA review. FDA’s 2016 guidance said routine cybersecurity updates and patches are generally treated as device enhancements for which FDA does not require advance notification or reporting. [3] ICS-CERT repeated FDA’s point that manufacturers can always update a device for cybersecurity, and that FDA does not typically need to review changes made solely to strengthen cybersecurity. [4]
The barrier was the manufacturer’s validation and the hospital’s contract. A device maker tests its software as part of the device. An operating system patch applied by the hospital could change that tested configuration. In our reading, a hospital that applies a patch itself may move the device outside the maker’s tested configuration and support terms, while the maker has to validate the patch on each model. FDA’s fact sheet, as ICS-CERT quoted it, also recognized that health care delivery organizations may need to patch or change devices and supporting infrastructure, and recommended working closely with manufacturers. [4]
Old operating systems in new equipment
The NAO described a longer history behind the device gap. The Department of Health and the Cabinet Office had written to trusts in 2014 offering temporary help with security for old equipment until April 2015, after which there would be no support. [5] Even so, the Department told the NAO that about 5 percent of the NHS IT estate, including computers and medical equipment, still ran Windows XP on May 12, 2017. [5] The NAO said this is partly because it is not always possible to remove or update Windows XP in applications and services built on it. [5]
That last point is the device gap in one sentence. When an operating system is part of a validated product, its end of support is not a date the hospital controls. The hospital inherits the maker’s choice of platform and the maker’s pace of moving off it.
What the contract said, or did not
The contract for a device decides who patches it, how fast, and who pays. In our reading, a contract silent on operating system updates leaves the question to negotiation after the fact. A hospital that had bought an imaging system with an embedded operating system had bought that operating system’s lifecycle too. When the maker stopped supporting the version inside, the hospital owned a device it could not secure and could not easily replace.
Reading a manufacturer’s notice
A manufacturer’s notice is only useful if the hospital can act on it. In our reading, a usable notice answers five things. Which products and versions are affected. Whether the maker has validated the operating system patch. How and by whom the patch is installed. What to do in the meantime. And how to reach the maker for site-specific help. A notice that answers fewer leaves the hospital to call the maker, product by product, during the incident.
The hospital’s side of that exchange needs its own records. It has to know which of its devices match the listed products and versions, where they are, and who in clinical engineering is responsible for each. Without those records, the maker’s notice cannot be matched to the floor.
Who pays for a device patch
A device patch has a cost, and someone bears it. The maker spends engineering time to validate. A field engineer may need to visit. The hospital may lose the device for a session while it is updated. In our reading, contracts that do not assign those costs invite delay, because each side waits for the other to commit. A hospital buying new equipment can settle this before signing: security updates included in the service agreement, at no separate charge, for a stated period.
What the notices changed
The notices turned an abstract risk into a list. Over the following weeks, makers issued customer notifications with recommendations, and ICS-CERT collected links to them in one place. [4] A hospital that kept an inventory of devices by maker and model could match the notices to its floor. One that did not had to send staff to find the equipment. In our reading, the clinical engineering team, which maintains medical equipment, became a security team that week whether or not it had been one before.
The network between them: containment in place of patching
The third part asks how hospitals protected machines they could not patch in time. US-CERT’s answer was to block SMB at the network boundary if the patch could not be applied, while weighing the disruption. [1]
Boundary and internal controls
The NAO reported a finding that bridges the first two parts. Whether organizations had patched their systems or not, taking action to manage firewalls facing the internet would have guarded organizations against infection. [5] NHS Digital confirmed that the ransomware spread via the internet, including through the N3 network connecting NHS sites. [5]
Boundary blocking protects against infection from outside. Inside the network, the dropper scanned local IP ranges and tried port 445 on each address. [1] An unpatched device on a flat internal network was reachable from any infected machine on it. Segmentation, meaning separate network zones with controlled traffic between them, limited that reach. In our reading, a device zone that accepted SMB only from the systems that needed it would have contained an outbreak that started elsewhere in the hospital.
Isolation and its cost
The NAO said trusts running Windows XP on medical equipment could have protected themselves by isolating the devices from the rest of the network, although this may require manual workarounds. [5] That trade is the core of the third part. An isolated scanner is safe and cannot send images to the clinicians who need them. [5] US-CERT made the same point about SMB: blocking it may obstruct access to shared files, data or devices. [1]
A hospital that had designed its network in advance could isolate a device class quickly and keep a controlled path for clinical data. A hospital that had not faced a binary choice: connected and exposed, or isolated and disruptive. In our reading, that binary choice is what turned a technical exposure into a clinical one.
Watching the kill switch domain
US-CERT advised administrators not to block traffic to the domain the malware checked. [1] That advice had a second use. A machine trying to reach the domain was a sign of infection, or of a security tool testing it. The NAO reported that between May 15 and mid-September, NHS Digital and NHS England identified a further 92 organizations, including 21 trusts, contacting the domain. [5] It noted some may have done so as part of security activity. [5]
In our reading, that shows containment and detection working together. Network logs that record outbound connections let a hospital find infected machines it did not know about, including devices it cannot inspect directly.
Precautionary shutdowns
The NAO found that some disruption came from trusts that were not infected. [5] Without timely central direction, some trusts shut down devices, isolated them or disconnected from the N3 network on their own initiative. [5] Forty-six trusts were not infected but reported disruption. [5] Those decisions were reasonable under uncertainty. They show that the containment decision has clinical cost even when no machine is encrypted.
What containment requires in advance
Containment works when four things exist before the incident. A map of which device classes sit in which network zones. Rules that can be tightened for one protocol without shutting a zone down. A clinical owner for each device class who can approve isolation. And a list of the data flows each device needs, so a controlled path can stay open. These cannot be built in the hours after an outbreak begins.
The clinical work on top: keeping care running
The fourth part asks what kept patients safe while machines were locked or isolated. The NHS account is the clearest record of that experience.
What happened to care
Infected trusts faced two kinds of disruption, the NAO said. Staff were locked out of devices, which delayed access to patient information, test results, transfers and discharges. [5] And medical equipment was locked or isolated, which disrupted radiology and pathology departments. [5] Some trusts used manual workarounds, recording information on paper. [5]
NHS England identified 6,912 appointments that had been canceled and estimated over 19,000 would have been canceled in total. [5] Of the 25 acute trusts infected and locked out, 20 continued treating urgent and emergency patients through the weekend. [5] Five diverted patients to other emergency departments, and two more needed outside help. [5] By May 16, only two hospitals were still diverting patients. [5]
Plans that had not been rehearsed
The NAO found the Department of Health had a plan with roles for responding to an attack, but had not tested it at a local level. [5] With no rehearsal for a national cyber attack, it was not immediately clear who should lead, and communications were difficult. [5] Many organizations could not use email because they were infected or had shut email down. [5] Staff shared information through personal mobile devices, including an encrypted messaging application. [5]
In our reading, those findings apply to any hospital. A downtime procedure written for a power failure assumes the computers will come back as they were. A ransomware event may lock some systems, leave others running but untrusted, and cut email. The procedure needs to say how staff communicate, how orders and results move, and who decides when a system is safe to reconnect.
Who decides to reconnect
Isolation is the easy decision in an outbreak. Reconnection is harder. A device or system comes back only when someone is satisfied it is patched or otherwise protected, and that no infection remains. In our reading, that decision needs a named owner for each system class, a short checklist, and a record. For a clinical device, the owner should include clinical engineering and the department that uses it. For a records system, it should include the clinical leadership that relies on it.
Reconnecting too early risks reinfection. Reconnecting too late prolongs paper work and its re-entry burden. A hospital that has agreed the criteria in advance avoids arguing them during the event.
Recovery after the event
The NAO said trusts and other NHS organizations had to roll back systems and restore data, including re-entering data recorded manually while systems were down. [5] It said the NHS had not calculated the total cost of canceled appointments, overtime, extra IT support or restoration. [5]
Re-entry is the hidden cost of paper downtime. Every order, result and note recorded on paper has to reach the electronic record, accurately, after the systems return. In our reading, a downtime procedure should say who does that work, how it is checked, and how clinicians know which records are complete. Otherwise the clinical risk moves from the outage to the reconciliation.
Earlier warnings
The NAO also reported that some trusts had been attacked before. Two of the trusts infected by WannaCry had been infected by previous cyber attacks. [5] One trust had suffered a ransomware attack in October 2016 that led to the cancellation of 2,800 appointments. [5] National reviews published in July 2016 had warned the Department that cyber attacks could jeopardize access to critical patient record systems. [5] The fourth part had been tested before, at smaller scale.
Downtime as a capability
A hospital that can run safely on paper for a defined period has bought time for the first three parts. It can isolate devices without stopping care. It can delay a reconnection until the patch is confirmed. In our reading, downtime capability is a security control as much as a clinical one. It changes the cost of containment, and so changes which containment decisions a hospital is willing to make.
How the week reached different health organizations
The four parts applied across health care. They played out differently by organization type. The descriptions below are our reading of how the structure fit each setting.
Large health systems
A large system has both gaps at scale: thousands of computers under its own control and thousands of devices from many makers. Its advantage is staff. It can run a fast lane for its own machines and a clinical engineering team for devices. Its risk is complexity. Several hospitals on one network, acquired over time, may not share a segmentation design. A system that acquires a hospital also acquires its devices, its contracts and its patch backlog. In our reading, a security review of the device inventory and network design belongs in the integration plan, before the new hospital joins the shared network. Otherwise the weakest site sets the exposure for all of them.
Community and rural hospitals
A small hospital may have a handful of IT staff and depend on a maker’s field engineer for each device class. For it, the second part is hardest. It cannot validate patches itself and may wait on several makers at once. Network isolation and downtime procedures carry more of the load.
Ambulatory networks and physician practices
Clinics run fewer large devices and more desktops, in our reading. For them the first part dominates. The NHS figures for GP practices show how fast a patch campaign could move across many small sites when it was coordinated. [5] A network of clinics with central IT can do the same. Independent practices depend on their IT providers.
Laboratories and imaging centers
Laboratories and imaging centers concentrate the device problem. Their core equipment is the kind the NAO described: vendor-managed, with embedded operating systems. [5] For them, the maker’s notice is the critical document and the maker’s support contract is the critical control.
Academic medical centers
Academic centers add research equipment and laboratory systems to the clinical fleet. Research instruments may run old software kept for reproducibility, and may sit on networks shared with clinical systems. In our reading, they face the device gap with less vendor support than clinical equipment, because research tools may not be regulated as medical devices at all. Segmentation between research and clinical zones carries more weight for them.
Device manufacturers
For makers, the week was a test of their own processes. Each had to determine which products used affected Windows versions, validate a fix, and tell customers. [4] FDA’s guidance had already said routine patches generally did not need advance notification to FDA. [3] The remaining constraint was each maker’s engineering and support capacity.
Public health and government health agencies
Public agencies running laboratories or clinical services face both gaps, under public procurement rules. Their device contracts may be long and hard to change. For them, the fourth part also covers public communication when a service is disrupted. A public laboratory that stops reporting results affects clinicians and patients outside its own walls. Its downtime plan has to say how results reach those who depend on them, and how the public is told about delays.
Objections, in order
1. WannaCry was a Windows XP problem
The NAO found the opposite for the NHS. The majority of infected devices ran supported Windows 7 without the March patch, and XP devices were a minority of identified issues. [5] XP mattered for embedded medical equipment that trusts could not update. [5] Both gaps were real. In the NHS, the larger one was the trusts’ own.
2. The kill switch ended the threat, so the week was over-dramatized
The kill switch slowed the spread. [1] [5] The NAO said the attack could have caused more disruption without it. [5] Other malware also exploited the same SMB flaws, which ICS-CERT reported in its alert. [4] The exposure remained until machines were patched or contained.
3. FDA rules stop hospitals patching devices
FDA’s guidance said routine cybersecurity patches generally do not need advance notification or reporting. [3] ICS-CERT repeated that manufacturers can always update a device for cybersecurity. [4] The practical constraints were the maker’s validation and the contract, not FDA review.
4. Isolating devices is always the safe choice
Isolation removes network exposure and can stop clinical data flows. [1] [5] The NAO recorded MRI results that could not reach clinicians. [5] Isolation is safe for the device and costly for care unless a controlled path is designed in advance. The right question is which data flows each device needs, and whether the network can keep those open while closing everything else.
5. US hospitals are different from the NHS
They are, in structure and payment. The two gaps are features of the technology, not the health system. A US hospital runs Windows computers it patches and devices it does not. [2] [4] The NAO’s account shows how those gaps behave under attack. [5] What differs is the governance around them: who can require a hospital to act, who pays for replacement, and who coordinates a national response. Those differences change the remedies, not the gaps.
6. Paying the ransom would have been faster
US-CERT said paying does not guarantee decryption and does not remove the malware. [1] The Department of Health, NHS England and the National Crime Agency told the NAO that no NHS organization paid the ransom. [5] Recovery depended on restoring and patching systems, which is the work the first three parts describe.
7. Faster patching would cause clinical outages of its own
It can, which is why the fast lane in the first part is narrow. It covers critical flaws on exposed services, with a reduced test focused on core clinical applications. The alternative was shown in the NHS: machines on supported Windows left unpatched despite two national alerts. [5] An outage from a bad patch can be reversed by rolling back. An outage from ransomware may not be.
8. Manufacturers will fix their own devices, so hospitals need not track them
Manufacturers issued notices. [4] Hospitals still had to match notices to equipment, decide on isolation and keep care running. [5] FDA’s later requirements put more duty on makers for new devices. [8] They do not remove the hospital’s need to know what it owns.
What hospital technology and security leaders could check
The four parts translate into checks a hospital can run today, for SMB or any flaw that reaches both its computers and its devices.
- For a critical flaw on an exposed service, how quickly can the hospital list affected machines and patch them, and who can authorize restarts?
- Does the device inventory record maker, model and embedded operating system, so a manufacturer’s notice can be matched to the floor in hours?
- Do device contracts state who applies security updates, on what timeline, and what happens when the embedded operating system loses support?
- Can the network isolate one device class from one protocol while keeping a controlled path for clinical data?
- Has the downtime procedure been rehearsed for an event where email is down and some systems are untrusted?
The checks also apply beyond SMB. Any flaw that reaches both the hospital’s own computers and its vendors’ equipment will split the same way. The hospital’s side is speed. The vendor’s side is contract and validation. The network decides the reach, and downtime decides the harm. A hospital that reviews its last incident against these four parts will find which one was weakest.
The second and third checks address the gap a patch metric hides. A hospital can report full compliance on the machines it controls and still carry a device fleet it cannot change. [5] The record of which is which is where the work begins.
With hindsight
The NAO published its investigation in October 2017. [5] Its account, used throughout this article, is a detailed official public record of a health system under WannaCry. It found that at least 80 of 236 trusts in England were affected, and that the Department of Health did not know the full cost. [5] It said NHS England and NHS Improvement had written to every trust, clinical commissioning group and commissioning support unit. [5] Boards were asked to make sure they had implemented all 39 CareCERT alerts NHS Digital issued between March and May 2017 and had taken essential action to secure local firewalls. [5]
The NAO also recorded follow-up on unsupported software. In July 2017, the Department responded to the National Data Guardian’s review. [5] As part of that response, it told local bodies to make sure they had moved away from, or were actively managing, unsupported software by April 2018. [5] That instruction matches the second part of our argument. It asked organizations to manage a risk they could not always patch.
The US government later attributed the attack. US-CERT updated its alert to reflect the government’s public attribution of the WannaCry variant to the North Korean government. [6] The alert’s current version was last revised on June 7, 2018. [6]
CISA later listed the flaws behind WannaCry as known exploited vulnerabilities. It added CVE-2017-0143 to the catalog on November 3, 2021. [7] It added CVE-2017-0144 and CVE-2017-0145, both SMBv1 remote code execution flaws, on February 10, 2022. [7] The catalog marks known ransomware campaign use as Known for all three. [7]
In our reading, the second gap drew the larger regulatory change. FDA says Congress granted it explicit cybersecurity authority in 2022 by adding section 524B to the Federal Food, Drug, and Cosmetic Act, effective March 29, 2023. [8] The section applies to premarket submissions for a cyber device. [8] FDA describes a cyber device as one that includes software validated, installed or authorized by the sponsor, can connect to the internet, and could be vulnerable to cybersecurity threats. [8]
Under section 524B, the sponsor must submit a plan to monitor, identify and address postmarket vulnerabilities and exploits in a reasonable time. [8] It must maintain processes to provide reasonable assurance that the device is cybersecure, and make postmarket updates and patches available. [8] It must provide a software bill of materials covering commercial, open-source and off-the-shelf components. [8] FDA’s premarket guidance asks for software bill of materials information including level of support and end of support for components. [8]
Those requirements answer the second part directly, for new devices. A maker must now plan, before clearance, how it will deliver patches and say which components it relies on and how long they are supported. [8] The requirements do not apply to submissions made before March 29, 2023. [8] Devices already in hospitals in 2017 were bought under the older terms.
What held up? Both gaps did. The hospital’s own patch speed remains a hospital responsibility, and the SMB flaws remained in CISA’s catalog with known ransomware use. [7] The device gap moved partly to manufacturers by law, for new submissions. [8] What did not hold up was the idea that WannaCry was mainly about old Windows. The NAO’s evidence showed supported, unpatched machines carried more of the exposure. [5]
Spotted an error? Tell us and we will check it and correct the article where needed.