BlueKeep asked organizations to patch on a forecast, before any attack

Microsoft warned in May 2019 that a Remote Desktop flaw was wormable, though no one had seen it exploited. Acting on that forecast took four things: criteria to judge it, a fast view of exposure, an emergency patch path and mitigations to hold the line.

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

Microsoft warned in May 2019 that a Remote Desktop flaw was wormable, though no one had seen it exploited. Acting on that forecast took four things: criteria to judge it, a fast view of exposure, an emergency patch path and mitigations to hold the line.

  1. The week On May 14, 2019, Microsoft released a fix for a critical remote code execution vulnerability, CVE-2019-0708, in Remote Desktop Services, the component formerly known as Terminal Services.
  2. Judging the forecast Several signals pointed the same way.
  3. Finding the exposure Once the flaw is declared an emergency, two questions decide the work: which systems run the affected software, and which of them accept connections to the vulnerable service from where.
  4. Moving at forecast speed Microsoft said that customers on in-support versions with automatic updates enabled were protected automatically.
  5. Holding the line For systems that could not be patched, CISA’s June 2019 alert listed mitigations besides upgrading: disable services the system does not use, enable Network Level Authentication on Windows 7, Server 2008 and Server 2008 R2, and block TCP port 3389, used to start Remote Desktop sessions, at the enterprise perimeter firewall.
  6. Two modes in one program The bulk of a patch program’s work is routine: monthly fixes deployed on a schedule, with priority raised for flaws known to be exploited.
  7. After the emergency An emergency is not over when the patch is deployed; it is over when the organization has checked that the fix is installed and the exposure is gone.
  8. Where the argument is weakest The strongest objection is that the forecast was wrong in its most alarming form.
  9. How it applies by sector Financial institutions may run Windows across branches, call centers, back offices and devices such as teller workstations and self-service terminals, and some of those systems may hold or process customer and payment data.
  10. With hindsight CISA issued an alert on BlueKeep on June 17, 2019.

The week

On May 14, 2019, Microsoft released a fix for a critical remote code execution vulnerability, CVE-2019-0708, in Remote Desktop Services, the component formerly known as Terminal Services. The flaw came to be called BlueKeep. The same day, the Microsoft Security Response Center published a warning about it that read more like a forecast than a bulletin. The vulnerability could be exploited before authentication and needed no user interaction. That made it “wormable”: future malware exploiting it could spread from vulnerable computer to vulnerable computer, as WannaCry had in 2017.

The company said it had observed no exploitation. It also said it was highly likely that malicious actors would write an exploit and incorporate it into their malware, and that affected systems should be updated as quickly as possible to prevent that scenario. The National Vulnerability Database, which published its entry on May 16, 2019, rates the flaw 9.8 out of 10 under version 3.1 of the Common Vulnerability Scoring System: reachable over the network, low complexity, no privileges and no user interaction required.

The affected systems were older releases of the operating system: the vendor named the Windows 7 and Server 2008 families among those still supported, and said the two newer releases, 8 and 10, were not affected. The Remote Desktop Protocol itself was not vulnerable; the flaw was in the service. The company took what it called the unusual step of also releasing fixes for XP and Server 2003, which were out of support, through a separate knowledge base article, KB4500705. Customers on supported versions with automatic updates enabled were protected automatically.

The vendor described a partial mitigation. With Network Level Authentication enabled, an attacker would need to authenticate before reaching the vulnerable code, which protected against wormable malware. A system with that setting remained exposed to an attacker holding valid credentials, so the company advised updating every affected system regardless.

The warning did not end that week. On May 31, 2019, Microsoft published a reminder. It said it was confident an exploit existed and that, if recent reports were accurate, nearly one million computers connected directly to the internet were still vulnerable, and that many more inside corporate networks might be too. It set out the timeline behind WannaCry: the MS17-010 fixes came out on March 14, 2017, a set of exploits including EternalBlue was published on April 14, 2017, and WannaCry began on May 12, 2017. It wrote that many customers had not patched in that period and that a significant number of them were infected.

For supported systems the fix shipped as an ordinary security update, through the vendor’s Security Update Guide and automatic updates. Organizations with automated patching for desktops could treat it as routine work at higher priority. Those whose servers were patched in scheduled change windows, or whose systems needed a vendor’s approval for each update, had to decide whether to wait for the next window or break it.

So in the week of May 13, 2019, organizations had a fix, a severity rating, a vendor’s prediction and no evidence of attacks. What they did next depended on whether their patching was built to act on a prediction.

BlueKeep is remembered as the worm that did not come in 2019. That memory skips over the harder test it set. In May 2019, the evidence that would tell a security team to drop everything, reports of attacks in progress, did not exist. What existed was a vendor’s forecast, an extreme severity score, and a reminder of what had happened two years earlier when organizations waited for proof. Teams had to decide whether to treat a prediction as an emergency.

A patch program handles a steady flow of fixes, each assigned a priority. It can raise the priority of flaws that attackers are known to be exploiting; CISA’s catalog of known exploited vulnerabilities, which later listed BlueKeep, exists to make that evidence easy to use. BlueKeep sat on the other side of that line. Waiting for exploitation evidence before acting would have meant, for a wormable flaw, acting after the worm had started.

Our argument is that BlueKeep showed organizations need a defined way to act on a credible forecast, separate from the routine patch cycle and from the response to known exploitation. Acting on a forecast has four parts, which do not overlap and together cover the work. First, judging the forecast: deciding that this flaw belongs in the narrow class that justifies emergency action before attacks. Second, finding the exposure: knowing quickly which systems are affected and how reachable they are. Third, moving at forecast speed: patching through a path faster than the routine cycle. Fourth, holding the line: reducing exposure on systems that cannot be patched at once. We take each in turn, then test the argument and look at how it applies to banks, health systems and public agencies.

This is a cross-sector argument because banks, health systems and public agencies all run Windows, and nothing about the affected versions was specific to one sector. The sectors differ in which systems are hardest to patch quickly and in what their regulators expect, and organizations should confirm with their own regulators and auditors how they will be judged on emergency patching.

Judging the forecast

The signals in May 2019

Several signals pointed the same way. The flaw could be triggered over the network, before authentication, with no user action, which is what the 9.8 score in the National Vulnerability Database records. The vendor called it wormable and said exploitation was highly likely. The vendor took the unusual step of fixing operating systems it no longer supported, a signal that it believed the risk was serious. And the vulnerable service, Remote Desktop, is a remote administration service that can be reachable from untrusted networks. Each signal alone might not justify an emergency. Together they described a narrow and recognizable class of flaw.

A short list of criteria

We think an organization should write down, in advance, the criteria that move a flaw from routine to emergency before any exploitation is reported. A workable list would include: remote exploitation without authentication or user interaction; a service that is commonly reachable from untrusted networks; a vendor or government statement that exploitation is likely or that the flaw is wormable; and a large population of affected systems in the organization. We think flaws meeting all of them are uncommon. Writing the criteria down lets a security team declare an emergency without arguing from first principles each time, and lets leaders see why it did.

Scores are an input, not a decision

A severity score describes a flaw in general. It does not know whether the organization runs the affected software, whether the service is reachable, or what the systems do. A high score alone may describe a flaw that never becomes an emergency for a given organization. The criteria above use the score as one input and add the facts that only the organization knows: how much of its estate is exposed and how much damage a spreading attack would cause.

Where forecasts come from

For BlueKeep, the forecast came first from the vendor, on the day of the fix. The National Vulnerability Database entry followed two days later, and CISA’s alert the next month. An organization that waited for the government alert before acting would have lost the time from May 14 to June 17. We think the criteria should name the sources a team watches on release day, with the vendor’s own advisories first, and should allow an emergency to be declared on the vendor’s statement alone when the other criteria are met.

Who declares the emergency

Someone has to decide, and the decision has costs: changes outside normal windows, staff pulled from other work, risk of disruption. We would give the chief information security officer authority to declare a forecast emergency against the written criteria, with immediate notice to the chief information officer and the business owners affected. A decision that needs a committee meeting will be late.

Recording a decision not to declare

Deciding that a severe flaw does not meet the criteria is also a decision, and it should be recorded with the same care. A short note saying which criteria were and were not met, who decided and when the decision will be revisited protects the team if the forecast later firms up. It also gives the after-action review something to test: whether the criteria drew the line in the right place. Over a few years, a file of such notes shows auditors and leaders that the organization weighs severe flaws deliberately, declaring some emergencies and declining others for stated reasons.

A forecast can be revised

A forecast is a judgment at a point in time, and new information can strengthen or weaken it. Publication of working exploit code, reports of scanning for the vulnerable service, or a government alert all move a forecast toward certainty. A vendor clarifying that a mitigation fully protects certain configurations can move it the other way. We would have the declaring officer revisit the decision at set points, and say so to leaders, so that the emergency tightens or relaxes as the picture changes rather than running on its first-day assumptions.

Telling leaders what the forecast is and is not

Executives asked to support emergency work on an unexploited flaw will reasonably ask how sure the team is. The honest answer in May 2019 was that no one knew when or whether an attack would come, but that the conditions for a fast-spreading one were present and the cost of being late was high. We think leaders should hear it that way, with the criteria that were met and the work that will follow, rather than as a certainty that would later look like crying wolf.

Finding the exposure

Two questions, fast

Once the flaw is declared an emergency, two questions decide the work: which systems run the affected software, and which of them accept connections to the vulnerable service from where. For BlueKeep, the answers began with the versions Microsoft named in May, Windows 7, Server 2008, Server 2008 R2, XP and Server 2003 (CISA’s June alert also listed Windows 2000 and Vista), and among them those with Remote Desktop enabled and reachable. An organization that could answer both within a day could plan. One that needed weeks of discovery lost the time the forecast gave it.

Exposure decides the order

Not every vulnerable system carried the same risk. A Windows 7 machine with Remote Desktop exposed to the internet was in a different position from one on an internal network with the service disabled. Ordering patch work by exposure, starting with systems reachable from the internet, then systems reachable from many internal hosts, then the rest, gets the most risk reduction from the first hours of effort. That ordering depends on knowing which systems are reachable from where.

Looking from outside

Microsoft’s May 31 reminder cited reports that nearly one million internet-connected computers were still vulnerable. A count like that is made from outside the network, and an organization can ask the same question of its own address ranges. A scan of the organization’s internet-facing addresses for Remote Desktop, repeated after each change, shows the exposure an attacker would see and catches systems the inventory missed.

Finding what the inventory misses

No organization knows every system it runs, so exposure work built on an inventory will miss something. The answer is to look for what the inventory misses: network scans for Remote Desktop and other remote access services, endpoint management reports of operating system versions, and reviews of vendor-supplied devices during procurement and maintenance. Each finding outside the inventory is both a risk to address and evidence about where the inventory process is weak.

Systems inside products

Some of the hardest systems to patch quickly are those that run Windows as part of a product sold by someone else: a diagnostic device, a building control system, a kiosk, an industrial controller. The organization does not choose when that product’s operating system is updated; the product’s maker does. For these systems, the fix may need the maker’s approval or its own release, and the organization’s exposure list should flag them so that the maker can be contacted on the first day. Procurement contracts that require makers to assess and approve critical operating system fixes within a stated time make that call more productive.

Laptops, remote workers and cloud machines

Exposure is not limited to servers in a data center. Laptops that leave the office, virtual machines in cloud accounts set up by individual teams, and test systems built for a project and never removed can all run affected versions with the service enabled. Cloud machines with public addresses deserve particular attention, because a security group opened for convenience can expose Remote Desktop to the whole internet. The exposure search should cover every cloud account the organization holds as well as the main network.

Other people’s exposure

Service providers that connect to the organization’s network, or that host systems on its behalf, have their own affected systems. A vulnerable machine at a provider with a trusted connection is a path into the organization. On the first day of a forecast emergency, we would ask key providers whether they are affected, what they are doing and when they expect to finish, and record the answers.

Moving at forecast speed

Automatic updates did the work

Microsoft said that customers on in-support versions with automatic updates enabled were protected automatically. For desktops managed that way, BlueKeep needed no special effort. The exposure lay with systems where updates were not automatic: servers patched in change windows, systems whose vendors had to approve each update, and machines outside central management. Those systems are where a forecast emergency spends its effort.

Testing without stalling

CISA’s alert repeated its usual advice to test patches before installation. Testing protects against updates that break critical systems, and in regulated settings it may be required. It can also become the reason a critical fix waits weeks. We would define in advance a faster path for critical, remotely exploitable flaws: a short test on representative systems, staged deployment starting with the most exposed, and a rollback plan. That path should be agreed with the business before it is needed, not negotiated during the emergency.

The window is set by attackers

Microsoft’s May 31 reminder used WannaCry to make a point about time. Fixes for the flaws exploited by WannaCry were released on March 14, 2017, exploits became public on April 14, and the attacks began on May 12. For BlueKeep, a public exploit module appeared in September 2019, according to Microsoft’s later analysis. Neither timeline is a promise about the next flaw. Both show that the time between a fix and its exploitation is set by attackers, and that an organization’s patching speed for critical, remotely exploitable flaws should be measured in days for exposed systems.

Out-of-band changes need pre-approval

Change management protects production systems, and in regulated settings it is part of the control framework auditors test. A forecast emergency needs changes outside normal windows. We would agree in advance with the change advisory function a standing pre-approval for forecast emergencies declared under the written criteria, covering patch deployment and the mitigations below, with records kept as for any change and a review afterward. That keeps the audit trail intact without waiting for the next scheduled window.

Out-of-support systems on the list

For BlueKeep, the fixes for Windows XP and Server 2003 arrived through a separate knowledge base article rather than the automatic route. Teams with such systems had to know they had them, retrieve the fix and install it by their own means. An emergency plan should treat out-of-support systems as a named workstream with its own owner, because they will not be swept up by the tools that handle supported systems.

Reboots and availability

Operating system fixes require restarts, and restarts interrupt services. For systems that run around the clock, such as clinical systems, payment switches and emergency dispatch, the emergency plan needs a way to restart in turn, using redundancy where it exists and short planned interruptions where it does not. The business owners of those systems should know in advance that a forecast emergency may require a restart at short notice, and should have agreed how it will be done.

Telling users what is happening

Emergency patching can produce unexpected restarts and brief outages. A short message to staff explaining that a critical security fix is being installed, and what they may notice, reduces help desk calls and the temptation to postpone updates. For systems used by customers or the public, the communications team should have a holding statement ready in case a restart causes visible disruption.

Measuring progress daily

A forecast emergency should be tracked like an incident, with a daily count of affected systems, patched systems, mitigated systems and systems still exposed, broken down by exposure tier. That count tells leaders whether the effort is working and tells the team where to push. It also creates the record that shows, afterward, how fast the organization moved.

Holding the line

The mitigations available

For systems that could not be patched, CISA’s June 2019 alert listed mitigations besides upgrading: disable services the system does not use, enable Network Level Authentication on Windows 7, Server 2008 and Server 2008 R2, and block TCP port 3389, used to start Remote Desktop sessions, at the enterprise perimeter firewall. CISA noted that blocking the port would also block legitimate Remote Desktop sessions and might not prevent unauthenticated sessions from inside the network. Microsoft had said that Network Level Authentication protected against wormable malware but not against an attacker with valid credentials.

Each of those mitigations reduced exposure without removing the flaw. Together they illustrate the logic of isolation: if a system cannot be fixed, limit who and what can reach the vulnerable part of it.

Isolation covers paths as well as ports

A port blocked at the perimeter stops attackers on the internet. It does not stop an attacker who has already reached the internal network, for example through a phishing email or a compromised vendor account. Microsoft’s May 31 reminder made the point: one vulnerable computer exposed to the internet could be a gateway into a corporate network, where malware could spread to other vulnerable systems. Isolation of an unsupported system therefore means restricting internal paths as well: which subnets can reach it, which accounts can log in, which services it exposes and which other systems it can reach in turn.

Remote access is the riskiest path

BlueKeep was a flaw in a remote access service. Unsupported systems that accept remote administration, whether through Remote Desktop or a vendor’s own tool, offer an attacker a direct route in. We would remove direct remote access to unsupported systems wherever possible, route any remaining access through a hardened jump host with multifactor authentication, and log each session. Vendor access in particular should be enabled only when needed and disabled afterward.

Credentials still open the door

Microsoft noted that Network Level Authentication left systems exposed to anyone with valid credentials. That makes the accounts able to log in to an unpatched system part of its protection. Until the fix is in, we would limit interactive logins to a short list of named administrative accounts, keep those accounts separate from everyday user accounts, remove shared and default accounts, and avoid trust relationships that let a compromise there reach the wider directory. An unpatched system that any domain user can reach by Remote Desktop is not protected, whatever the firewall says.

Turning off what is not needed

CISA’s advice included disabling services the operating system does not use. On many workstations and some servers, Remote Desktop may be enabled without a current business need, left over from a setup or a past support session. Turning it off removes the exposure entirely for those systems, with no patch and no restart. We would make that check part of the first day’s work, starting with systems where no one can name a reason for the service to be on.

Watching for the first attempts

A system waiting for its fix should be watched more closely than one that has it. Monitoring for unexpected connections, new processes and crashes gives early warning. Microsoft’s November 2019 analysis of the first BlueKeep attacks reported a rise in Remote Desktop service crashes that it linked to the unstable exploit module. An organization watching its unsupported systems for that kind of signal would have had a chance to notice exploitation attempts.

Trading speed for protection

For some systems, the safest course is a slower, carefully tested patch with strong mitigation in the meantime. A clinical system, a payment switch or an industrial controller may justify that trade. The trade should be explicit: which mitigations make the slower path acceptable, how long the system will wait for its fix, and who agreed. Without that record, a slow patch looks the same as a forgotten one.

Mitigations need an end date

Mitigations that block a service or a port can outlive the emergency, or be reversed when they get in someone’s way, and either outcome causes trouble later. We would record each temporary mitigation with the system it covers, the reason and the condition for removing it, such as the fix being installed and verified. Reviewing that list as the emergency closes keeps temporary measures from becoming permanent gaps or silent reversals.

Two modes in one program

How the modes fit together

The bulk of a patch program’s work is routine: monthly fixes deployed on a schedule, with priority raised for flaws known to be exploited. The forecast mode we describe is an exception to that schedule, reserved for a few flaws. The two modes share the same inventory, tools and people. What differs is the trigger and the speed. We think that framing matters because it keeps the forecast mode from becoming a parallel program with its own staff and budget, and keeps the routine program from absorbing every alarm as an emergency.

When a forecast becomes evidence

A flaw handled as a forecast emergency does not leave the program when the emergency closes. Systems added later, restored from old images or found in a forgotten corner can bring the flaw back. If attackers later begin to exploit it, the flaw will appear on government lists of known exploited vulnerabilities, and any remaining exposure becomes a routine-mode priority with evidence behind it. We would keep forecast-emergency flaws on a watch list, checked in each routine scan, so that the evidence mode catches what the forecast mode left behind.

Capacity for the unexpected

A program that runs at full capacity every month has no room for a forecast emergency. Leaders who want the organization to move fast on the next BlueKeep should accept that some capacity must be held in reserve, or that planned work will be delayed when an emergency is declared. Deciding which planned work gives way, in advance, saves an argument in the middle of the emergency.

After the emergency

Verifying the fix

An emergency is not over when the patch is deployed; it is over when the organization has checked that the fix is installed and the exposure is gone. That means scanning again, internally and from the internet, and reconciling the results with the deployment records. Systems that report success but still respond as vulnerable, and systems that were missed entirely, show up only in that check.

Closing the temporary measures

Once fixes are verified, the temporary mitigations need a decision. Some should stay, such as blocking Remote Desktop at the perimeter or turning off a service no one needs. Others, such as emergency firewall rules that disrupt legitimate work, should be removed in a controlled way. Each decision should be recorded so the next team knows why a rule exists.

Reviewing the decision

A short review after each forecast emergency should ask whether the criteria fitted, how long each part took, which systems were hardest and why, and what would make the next one faster. Over time, those reviews show whether the organization is getting better at acting on forecasts and where its inventory, change process or provider arrangements are holding it back.

Reporting to the board

Boards and senior leaders hear about incidents. They may never hear about attacks that did not happen because a team moved early, which makes forecast emergencies hard to fund. We would report each forecast emergency briefly: what the forecast was, what was done, how fast, and what was learned. Those reports build the record that justifies the capacity a forecast process needs.

Where the argument is weakest

The worm did not come in 2019

The strongest objection is that the forecast was wrong in its most alarming form. No worm swept the internet in 2019. Organizations that treated BlueKeep as routine did not suffer a WannaCry-style outbreak from it that year. We accept that. Forecasts are judged by whether the decision was sound given what was known, not by whether the worst case arrived. The conditions Microsoft described were real, and as the hindsight below shows, attackers did exploit the flaw within months.

Crying wolf has a cost

A second objection is that emergencies declared on forecasts that do not come true wear out the people and the goodwill they depend on. That is a fair concern, and it is why the criteria need to be narrow and written down. An organization that declares a forecast emergency every month has criteria that are too loose. One that declares one a year, for flaws like BlueKeep, is spending its credibility where it matters.

Known exploitation is a better signal

A third objection is that patch priority should follow evidence of exploitation, since a severe flaw may never be exploited. For the bulk of fixes we agree, and government lists of exploited vulnerabilities have since made that evidence easy to use. The exception is the narrow class of flaws where exploitation, once it starts, spreads faster than an organization can respond. For a wormable flaw in an exposed service, the evidence arrives too late to act on. A patch program needs both modes: evidence-driven priority for the many, forecast-driven emergency for the few.

Emergency changes carry their own risk

A fourth objection is that rushing patches onto production systems can break them, causing the outage the patch was meant to prevent. That risk is real for critical systems. It is the reason for a fast test path, staged deployment and rollback plans, and for mitigations that reduce exposure while a cautious patch proceeds. The choice is not between rushing and waiting; it is between a prepared fast path and an improvised one.

Small teams cannot run two modes

A fifth objection is that small organizations barely keep up with routine patching and cannot maintain a separate emergency mode. We think the forecast mode can be small: a one-page set of criteria, a named person who can declare, a pre-approved change route and a list of exposed systems. For a small team, the most valuable part may be the first-day check for services exposed to the internet, which can be done with modest tools and removes much of the risk a wormable flaw presents.

How it applies by sector

Banks and credit unions

Financial institutions may run Windows across branches, call centers, back offices and devices such as teller workstations and self-service terminals, and some of those systems may hold or process customer and payment data. For them, the hardest part of a forecast emergency may be devices managed by outside providers. We think an institution should be able to show its examiners and auditors how it identified affected systems, including those its providers run, how fast it acted and what protected systems in the meantime.

Institutions also depend on service providers whose systems they do not see directly, such as core processors and card processors. A forecast emergency should include a same-day inquiry to key providers about their exposure and plans, recorded as part of vendor oversight. The same inquiry, with the answers and dates, is evidence an examiner can review afterward, and it shows the institution treated its providers’ exposure as its own.

Health systems

Hospitals depend on clinical devices and departmental systems that may run older Windows inside a product they did not build. Patching those devices can require the maker’s involvement, and taking them offline affects care. For health systems, holding the line carries more of the load: segmenting devices, cutting remote access paths and watching for exploitation while makers validate fixes. Clinical engineering and information security need a shared list of devices and a shared process so that the first day of a forecast emergency is spent acting, not discovering.

Health systems also connect to many outside parties: affiliated practices, laboratories, imaging centers and vendors with remote support access. A wormable flaw can travel along those connections. A forecast emergency in a health system should include a check of which outside connections reach affected systems, and whether they can be narrowed until the fixes are in.

Public agencies

Agencies operate many systems on funding cycles that do not allow for surprise work, and some run legacy applications on older platforms. A forecast emergency pulls staff from planned work; agencies should be able to account for that time and its effect on other priorities. Smaller governments, such as towns, counties and special districts, may have very little staff to manage such work, so isolation and remote access controls may be their realistic first steps, with help from state programs.

Agencies that deliver public safety and emergency services face the restart problem in its sharpest form, because their systems cannot go down at a convenient hour. For them, the plan for a forecast emergency should name, in advance, how each such system will be restarted and who approves it, so that a critical fix does not wait for a maintenance window that never comes.

What a leader would have asked in May 2019

Across sectors, the questions a CIO or CISO could have put to their teams after May 14 were the same: whether BlueKeep met the organization’s criteria for acting before attacks; how many systems ran the affected versions, and which accepted Remote Desktop connections from the internet or from wide internal networks; how fast the fix could be deployed to those first; and what would protect the systems that could not be patched that week. An organization that could answer quickly had a forecast process, whether or not it called it one. One that could not had a gap the next wormable flaw would find.

Those questions did not require knowing whether a worm would come. For that reason they were the right ones. They turned an uncertain forecast into a bounded piece of work with an owner, a measure of progress and an end, which is what a busy leadership team needs in order to support acting before the evidence arrives.

With hindsight

CISA issued an alert on BlueKeep on June 17, 2019. It said it had coordinated with external stakeholders and determined that Windows 2000 was also vulnerable, and it listed Windows Vista among the affected versions. Its mitigations were the ones we described: install patches, upgrade end-of-life operating systems, disable unnecessary services, enable Network Level Authentication and block TCP port 3389 at the perimeter, with a caution that blocking the port might not prevent unauthenticated sessions started inside the network.

The forecast was half right. Exploitation came later in 2019, and not as a worm. Microsoft’s security team wrote on November 7, 2019, that it saw an increase in Remote Desktop service crashes starting on September 6, 2019, when a BlueKeep exploit module for the Metasploit penetration testing framework was released, and crashes on researchers’ honeypots from October 23. It linked the activity to a campaign that installed a coin miner, and said the attacks were likely initiated as scans for machines with vulnerable internet-facing Remote Desktop services. It said there were no other verified attacks involving ransomware or other malware as of that writing, but that the exploit would likely be used to deliver more damaging payloads and that BlueKeep would remain a threat as long as systems stayed unpatched.

The flaw later moved into the evidence-driven world. CISA’s Known Exploited Vulnerabilities catalog, in the version dated November 13, 2026, includes CVE-2019-0708 with a date added of November 3, 2021, and a required action to apply updates per vendor instructions. Its field for known use in ransomware campaigns reads “Known.” For any organization still exposed by then, the forecast of May 2019 had become a listed fact.

Read together, those sources support both halves of our argument. Organizations that acted on the forecast in May 2019 had the nearly four months between May 14 and September 6 to close the door before the exploit module Microsoft described appeared. Those that waited for evidence got it, first as crashes and coin miners and, by the catalog’s current version, as a listing that marks known ransomware use. A patch program built only on known exploitation would have done the right thing for BlueKeep eventually, but late. A program with a narrow, written route for forecast emergencies would have acted when the conditions were clear.

The flaw also stayed in the record of advice. CISA’s June alert remains on its site as archived content, and Microsoft’s May posts remain on its security response blog, so the reasoning available to organizations in May 2019 can still be read today in the words used at the time.

What held up is the case for having two modes: evidence-driven priority for the bulk of fixes, and forecast-driven emergency for the rare flaw that is remote, unauthenticated, wormable and widely exposed. What did not hold up was the most alarming version of the forecast, a worm on the scale of WannaCry, at least in 2019. We do not think that weakens the case. The decision in May 2019 had to be made on what was known then, and what was known then justified acting.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • BlueKeep
  • CVE-2019-0708
  • patch management
  • Remote Desktop
  • vulnerability prioritization