An exploited kernel flaw in an older platform showed that platform age is a control patches cannot supply
Microsoft’s March 2019 fix closed a kernel flaw Google believed might be exploitable only on Windows 7, because newer versions carried mitigations. Patch compliance reports could not show that gap; a platform register beside them can.
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’s March 2019 fix closed a kernel flaw Google believed might be exploitable only on Windows 7, because newer versions carried mitigations. Patch compliance reports could not show that gap; a platform register beside them can.
- The week On Tuesday, March 12, 2019, Microsoft released its March 2019 security updates.
- The chain: a browser flaw and a kernel flaw Google’s description of the attack is worth reading as a lesson in how layered defenses fail together.
- Mitigations as controls an old platform cannot receive The central point in Google’s post was a single sentence: the vulnerability “may only be exploitable on Windows 7 due to recent exploit mitigations added in newer versions of Windows.”
- Where old platforms persist Old platforms persist for reasons, and the reasons sort into a few categories, each needing its own plan.
- The interval with no patch The days between Google’s disclosure and Microsoft’s fix were short, but they show the shape of a problem that recurs.
- The bridge: what extended updates buy When a platform reaches the end of its support, the vendor stops shipping fixes for it.
- Objections Five objections deserve answers.
- A platform register next to the patch report Pulling the parts together, we think an organization’s monthly vulnerability report should carry a second table beside patch compliance.
- With hindsight The deadline the bridge was built for arrived on schedule.
The week
On Tuesday, March 12, 2019, Microsoft released its March 2019 security updates. Two of the vulnerabilities fixed that day were elevation of privilege flaws in Win32k, a kernel component, and Microsoft’s own records marked both as exploited. The same day, the Cybersecurity and Infrastructure Security Agency posted an alert noting that Microsoft had released updates to address multiple vulnerabilities, some of which a remote attacker could exploit to take control of an affected system, and encouraged users and administrators to review the March 2019 Security Update Summary and apply the necessary updates.
One of the two, CVE-2019-0808, had been disclosed in public before the fix; Microsoft’s record marked it publicly disclosed and credited a researcher of Google’s Threat Analysis Group. On March 7, Google’s Threat Analysis Group published a post explaining that on Wednesday, February 27, it had reported two previously unknown vulnerabilities, one in Google Chrome and one in Microsoft’s operating system, that were being exploited together. Google had fixed the Chrome flaw, CVE-2019-5786, in an update for all Chrome platforms on March 1, pushed through Chrome’s automatic update, and asked users to verify they were on version 72.0.3626.121 or later.
The Windows flaw was different from the browser one. Google described it as a local privilege escalation in the win32k.sys kernel driver that could be used as a security sandbox escape, caused by a NULL pointer dereference under specific circumstances. Google wrote that it strongly believed the vulnerability “may only be exploitable on Windows 7 due to recent exploit mitigations added in newer versions of Windows,” and that it had observed active exploitation only against Windows 7 32-bit systems. It said it was disclosing the flaw before a fix because it was a serious vulnerability being actively exploited in targeted attacks, and that the unpatched flaw could still be used to elevate privileges or combined with another browser vulnerability to evade security sandboxes. As mitigation, it advised users to consider upgrading to Windows 10 if they were still running an older version, and to apply Microsoft’s patches when they became available.
Microsoft’s March release listed CVE-2019-0808 as affecting Windows 7 Service Pack 1, in 32-bit and x64 versions, and several editions of Server 2008 and Server 2008 R2. Its exploitability entry recorded the flaw as publicly disclosed and exploited, with exploitation detected on older software releases and “N/A” for the latest release.
The second exploited Win32k flaw, CVE-2019-0797, was a separate vulnerability. Microsoft listed it as affecting releases from 8.1 through the 10 series and Server 2012 through Server 2019. It was marked exploited but not publicly disclosed. Its exploitability entry read “Exploitation Unlikely” for the latest software release and “Exploitation Detected” for older releases.
So in one week, two exploited kernel flaws were fixed, one confined to Windows 7 and its Server 2008 siblings, and one reaching newer versions but rated by Microsoft as unlikely to be exploited on the latest release. With Google’s Chrome update of March 1 and Microsoft’s release of March 12, both links of the chain Google had described now had fixes, for machines whose owners applied them. Both flaws pointed to the same variable: which operating system release a machine was running, as well as whether it was patched.
Security teams measure patching closely. They count missing updates, track time to patch and report the share of endpoints that are current. The March 2019 Windows flaws showed the limit of that measure. A Windows 7 machine patched on March 12 was current. It was also running a version of Windows that, on Google’s reading, lacked exploit mitigations that would have blocked the attack on newer versions. Patching closed this particular flaw. It did not give the old system the protections that had made the newer ones harder targets.
Our argument is that an operating system’s version is a security control in its own right, separate from its patch level, and should be governed as one. Exploit mitigations are built into a platform release. They arrive when the platform is upgraded and cannot be patched into an older one. An organization that counts patches but not platform versions sees half of its exposure. We break the argument into seven parts:
- The chain: how the March 2019 attack combined a browser flaw and a kernel flaw, and how differently each was fixed.
- Mitigations as controls: why protections built into a newer platform are a control an older one cannot receive through updates.
- Where old platforms persist: the kinds of systems that stay on older versions, and why each needs its own plan.
- The interval with no patch: what an organization can do between disclosure and fix, and what it needs to know to do it.
- The bridge: paid extended updates, what they buy and what they do not.
- Objections: the strongest arguments against treating platform age as a control.
- The register: a platform register next to the patch report, and who acts on it.
The parts follow the logic of a decision. The opening two explain the exposure, the third locates it, the fourth covers the days before a fix, the fifth prices the delay, the sixth tests the argument, and the last turns it into a record and a routine.
The chain: a browser flaw and a kernel flaw
Google’s description of the attack is worth reading as a lesson in how layered defenses fail together. The browser runs web content inside a sandbox, a restricted process meant to stop a compromised page from reaching the rest of the system. The Chrome flaw, which CISA’s catalog later described as a heap use-after-free that a crafted HTML page could trigger, was the way in. The Windows kernel flaw, in Google’s words, could be used as a security sandbox escape, by elevating privileges. Each flaw alone was less useful. Together they gave an attacker a path from a web page to control of the machine.
Two fixes, two speeds
The two vendors fixed their halves differently. Google released a Chrome update on March 1 and pushed it through automatic update, so a browser that was allowed to update would have closed its half of the chain without any action by an administrator. Microsoft fixed the Windows half in its scheduled monthly release on March 12. Between those dates, Google told the public on March 7 that the Windows flaw existed, was exploited, and was not yet fixed.
That sequence matters for how an organization reads its own exposure. In a managed environment, browser updates may be held back for testing, and operating system patches may be staged across weeks. Each delay is reasonable on its own. But a chain attack needs only one open link. If the browser is current and the kernel is not, the chain may still work with a different browser flaw, as Google warned. If the kernel is current and the browser is not, the attacker has a foothold in the sandbox and can wait for the next escape.
What the disclosure asked of defenders
Google’s public disclosure before a fix put defenders in an unusual position. They knew a Windows flaw was being exploited, they knew which platform it affected, and they had no patch. Google’s advice for that interval was to consider upgrading to Windows 10. For an organization with a large Windows 7 estate, that advice could not be followed in days. The practical options were narrower: make sure Chrome had updated, reduce exposure of Windows 7 machines to untrusted web content where possible, and prepare to deploy Microsoft’s fix as soon as it shipped.
We think the interval exposed a planning gap. An organization that knew which of its machines ran Windows 7 32-bit, which of those were used for web browsing, and who used them, could act within hours. One that knew only how many machines were missing patches could not, because on March 7 no patch was missing. The flaw had no fix to be missing.
Automatic update as a design decision
The Chrome half of the chain was closed by automatic update. That was a design decision by the browser vendor, and it shaped how fast the fix reached users. An organization that manages browser updates centrally makes a different decision: it trades speed for control, testing each update before release. Both choices are defensible. What matters is that the organization knows which one it has made, and has a faster path for updates that fix exploited flaws.
We think the March 2019 sequence argues for a written rule: when a vendor says a browser flaw is being exploited, the update goes out on an emergency track, with testing reduced to what can be done in a day. The same rule can apply to operating system fixes for exploited flaws. Without such a rule, each exploited flaw becomes a fresh negotiation between the security team and the teams that own the test process, and the negotiation itself takes time an attacker can use.
Two vendors, one chain
The chain also crossed a vendor boundary. Google found both flaws and fixed its own. Microsoft fixed the other on its own schedule. Neither vendor controlled the whole attack, and neither could close it alone. An organization that tracks vulnerabilities vendor by vendor may see two unrelated items, one closed and one open, when the attacker sees a single path. Linking related items in the vulnerability record, so that the closure of one does not hide the openness of the other, is a small change with a clear benefit.
Mitigations as controls an old platform cannot receive
The central point in Google’s post was a single sentence: the vulnerability “may only be exploitable on Windows 7 due to recent exploit mitigations added in newer versions of Windows.” The flaw was a NULL pointer dereference in a kernel driver. On newer versions, Google said, recent exploit mitigations stood in the way. On Windows 7, they were absent.
Patches fix bugs; platforms change classes of bugs
A security update fixes a specific flaw. An exploit mitigation changes the conditions under which a whole class of flaws can be exploited. The difference matters for planning. The next Win32k bug on Windows 7 would need its own patch, and until that patch arrived, the platform would offer the attacker the same conditions as before. On a platform with the mitigation, the same class of bug might be far less useful to an attacker even before a patch.
Microsoft’s ratings for the other exploited flaw that week, CVE-2019-0797, point the same way. That flaw reached newer versions of Windows, including Windows 10 and Windows Server 2019. Microsoft rated exploitation of the latest software release as unlikely and recorded exploitation detected on older releases. We do not know from the records which mitigations explain that difference, and we do not claim to. But in both of the week’s records, the older releases were where the vendor had detected exploitation.
What an upgrade carries besides fixes
A platform upgrade may be budgeted as a compatibility project: will the applications run, will the users accept the change, how long will the rollout take. The security case may be reduced to the end of support date. The March 2019 flaw suggests a fuller case. The upgrade brings fixes the old platform will never receive, but it also brings protections designed after the old platform was built, which apply to flaws not yet found. That second benefit is harder to quantify, because it is about attacks that do not happen. It is still part of what the organization is buying.
We would ask a migration business case to state that benefit clearly, with the evidence available: vendor descriptions of the protections in the newer platform, and cases like this one where a flaw was exploitable on the old platform and, on the reporter’s belief and the vendor’s records, not on the new one. That gives decision makers a security argument for moving that does not depend on waiting for the support deadline.
A control that does not show up in patch metrics
If mitigations are controls, they need to be measured as controls. A patch compliance report treats a fully patched Windows 7 machine and a fully patched Windows 10 machine as equal. On the evidence of March 2019, they were not equal. The report would show both as compliant while one of them was exposed to an exploited flaw in a way the other was not.
We would add a second measure alongside patch compliance: platform currency. For each class of system, record the operating system version, the vendor’s published end of support date, and whether the version is the current release or an older supported one. A machine can then be patched and current, patched and on an older version, or unpatched. The second category is where the March 2019 attack lived.
Thirty-two bits
Google noted that it had observed exploitation only against Windows 7 32-bit systems. That detail is easy to pass over, but it is a reminder that platform version includes architecture. Google did not say why exploitation was confined to 32-bit systems, and we do not speculate. A 32-bit installation may also have been kept for a reason, such as an old application or driver that would not run otherwise. An inventory that records “Windows 7” without the architecture cannot answer the question Google’s post raised.
Where old platforms persist
Old platforms persist for reasons, and the reasons sort into a few categories, each needing its own plan. We would group them in four.
General-purpose desktops behind schedule
The simplest category is ordinary desktops and laptops that have not yet been migrated because the migration is a large project and has not reached them. These machines run standard office software and a browser. They are the easiest to move, and the main constraint is project capacity. The March 2019 flaw was a direct argument for moving browsing-heavy users first, because the chain started in the browser.
Machines tied to an application
The second category is machines kept on an old version because an application requires it: a line-of-business system certified on one version, a specialized client, or a driver for a peripheral. Banks may have branch or back-office applications in this position; health systems may have departmental clinical or imaging software; agencies may have case management clients or systems tied to specialized hardware. We cannot say how many such machines any sector had, and we do not suggest a number. But the category exists wherever an application vendor controls the upgrade timetable, and moving these machines depends on that vendor.
Devices that contain an operating system
The third category is equipment that runs a general-purpose operating system inside it: kiosks, point-of-sale terminals, building controls, medical devices, laboratory instruments and industrial systems. Microsoft’s own lifecycle page for Windows 7 lists Professional for Embedded Systems and Ultimate for Embedded Systems editions, a sign that the platform was built into equipment as well as desktops. Here the organization may not be allowed to change the operating system at all without the equipment maker, and the plan is about isolation, monitoring and replacement timing rather than migration.
Servers
The fourth category is servers. CVE-2019-0808 affected Windows Server 2008 and 2008 R2 as well as Windows 7. Servers may not be used for web browsing, in which case the browser-to-kernel chain is less relevant to them. But a kernel privilege escalation is useful to an attacker who already has a foothold by any route, and old servers may host applications whose owners are hard to find. We think servers on an old platform need the same platform currency measure as desktops, with the added question of who owns the application each one runs.
Finding them
Before any of these groups can be managed, they have to be found. Endpoint management tools can report the operating system of machines they manage, but the machines that matter for this argument are the ones those tools may not reach: devices supplied and maintained by vendors, systems on isolated networks, servers built for a single project and never enrolled, and equipment whose operating system is hidden behind a vendor’s interface. Network discovery, purchasing records and the asset lists kept by facilities, clinical engineering or branch operations can fill in the gaps.
We think the search is worth doing even when it is incomplete. A register that lists the old platforms an organization knows about, with an honest note of the areas not yet searched, is more useful than a confident total that leaves out the vendor-managed estate. The note itself tells leaders where the next search should go.
Why the categories matter
Each category has a different owner and a different lever. Desktops behind schedule belong to the IT migration program. Application-bound machines belong to the application owner and the vendor manager. Embedded devices belong to the equipment owner, and in health care, to clinical engineering. Servers belong to whoever owns the hosted application. A single figure for “machines on Windows 7” mixes all four and gives none of the owners a clear task. Splitting the figure by category turns it into four work plans.
The interval with no patch
The days between Google’s disclosure and Microsoft’s fix were short, but they show the shape of a problem that recurs. A flaw is known, exploited and unpatched. Patch management has nothing to do. The organization’s options come from other controls, and those controls are only as good as the information behind them.
Reduce exposure by platform
The first option is to reduce the exposure of the affected platform. If the chain begins in a browser, machines on the vulnerable platform can be steered away from untrusted web content for the duration: restricted to internal sites, routed through a remote browsing service, or limited to a list of allowed destinations. Each of these is disruptive, which is why they are reserved for short periods. Each also depends on knowing which machines are on the vulnerable platform. An organization that cannot list them cannot target the restriction, and may have to choose between restricting everyone and restricting no one.
Watch for the second stage
The second option is detection. A privilege escalation leaves traces: a process that should run with limited rights suddenly running with system rights, or a browser process spawning something it should not. Endpoint monitoring tools can be tuned to watch for those signs on the affected platform while the flaw is open. We do not suggest that monitoring replaces the fix. It shortens the time an attacker can use the gap, and it gives the organization evidence of whether the gap was used.
Decide who decides
The third option is organizational. Someone has to decide, quickly, whether to restrict browsing on a group of machines, accept the residual risk, or move users who browse untrusted sites to other equipment. That decision crosses security, IT operations and the business units that use the machines. If no one has the authority to make it on the day, the interval passes without a decision. We would name, in advance, the role that can make exposure-reduction decisions for an exploited and unpatched flaw, with a short list of pre-approved actions. That turns a disclosure like Google’s into a set of steps rather than a meeting.
All three options depend on the same record: which machines run which platform, used by whom, for what. The interval is when a platform register stops being a reporting nicety and becomes the basis for action.
The bridge: what extended updates buy
When a platform reaches the end of its support, the vendor stops shipping fixes for it. Microsoft’s lifecycle page lists the end of extended support for Windows 7 in January 2020. For organizations that could not move in time, Microsoft offered Extended Security Updates, available through specific volume licensing programs for Windows 7 Professional, Enterprise and Professional for Embedded Systems, for up to an additional three years past the end of support. The lifecycle page lists the third year of Windows 7 extended updates ending in January 2023.
What the bridge is, on the vendor’s terms
Microsoft’s own description of the program is precise about its limits. Its lifecycle FAQ calls Extended Security Updates “a last resort paid option” for customers who need to run certain legacy products past the end of support. It says they are not intended as a long-term solution but as a temporary bridge to stay secure while migrating to a newer, supported platform, and that they include security updates rated critical and important for up to three years after the end of extended support.
Read against March 2019, the bridge has a clear shape. It keeps fixes flowing for specific flaws in the old platform. It does not add the mitigations of the newer one. A Windows 7 machine on extended updates would receive a fix for the next Win32k flaw. It would still lack whatever newer versions had done to make that class of flaw harder to exploit. The bridge pays for patch currency. It does not buy platform currency.
Pricing the delay honestly
We think the decision to buy extended updates should be made with that distinction in view. The cost of delay is the license fee, plus the residual exposure that patches cannot remove, plus the risk that the migration slips again when the bridge ends. An organization that records only the fee will see delay as cheaper than it is. One way to make the residual exposure visible is to record, for each machine on extended updates, its category from the previous section, its owner, and the date by which it must move. The bridge then becomes a dated commitment with named owners rather than an open-ended subscription.
Measuring the bridge
Once an organization is on extended updates, it needs a measure of whether the bridge is being used as intended. We would track three numbers each quarter: how many systems remain on the bridged platform, by category; how many have a confirmed migration date before the bridge’s final year ends; and how many have no plan at all. The first number should fall. The second should rise toward the first. The third should reach zero well before the final year. If the numbers do not move, the bridge is becoming a destination.
The same measure helps with the renewal decision. An organization deciding whether to buy another year can see exactly which systems the purchase is for, who owns them, and why they have not moved. A renewal justified by a short list of named systems with dates is a different decision from a renewal justified by an estimate of how many machines might still need it.
The bridge as a deadline
Extended updates also change the shape of the deadline. Without them, end of support is a cliff. With them, it becomes a series of annual decisions. That can help, because it gives application vendors and equipment makers time to certify a newer platform. It can also hurt, because each annual renewal is a chance to defer again. In our view the bridge works well when the organization decides in advance which year it will be the last, and plans its migration to finish before then.
What the bridge cannot cover
Extended updates also have edges beyond the platform itself. They cover the operating system. They do not cover applications that their own vendors have stopped supporting on that platform, browsers that have ended support for it, or drivers and agents that no longer receive updates for it. A machine on extended updates may receive a kernel fix while running a browser or security agent that has stopped receiving any. The March 2019 chain began in the browser, so a bridged platform with an unsupported browser would be exposed at the first link of a similar chain, whatever its kernel patch level. We would check, for each bridged group, the support status of the browser and the security tools on it, and treat any unsupported component as part of the same decision.
Objections
Five objections deserve answers.
A patched system is a patched system
The first objection is that once Microsoft shipped the fix, the Windows 7 exposure to CVE-2019-0808 was closed, and the argument about mitigations is about hypothetical future flaws. That is true for the specific flaw. But the next flaw in the same class would reopen the exposure, and the window between exploitation and fix would again belong to the attacker. Google’s post described an exploit that was in use before the flaw was publicly known. Mitigations matter in exactly that window, when no patch exists. A control that works when patches cannot is not hypothetical.
Google’s statement was a belief, not a finding
The second objection is that Google said it “strongly believe[d]” the flaw might be exploitable only on Windows 7. It did not claim proof. We accept that and have quoted it as written. Microsoft’s own product list for CVE-2019-0808 named Windows 7 and Windows Server 2008 and 2008 R2, and no newer version. The vendor’s list and the reporter’s belief agree, which is enough for the argument without overstating either.
Upgrades bring their own risk
The third objection is that migrating a large estate carries its own operational risk: application breakage, user disruption and project cost. That is fair. Our argument is not that every machine should move at once. It is that platform age should be recorded and governed as a control, so that the organization makes the trade-off knowingly. A machine that stays on an old platform for a good reason, with an owner, compensating controls and a date, is a managed risk. One that stays because no one counted it is not.
Some equipment cannot move
The fourth objection is that some devices cannot be upgraded at all within their working life, because the maker will not support a new operating system on them. That is the case for the embedded category, and the answer for those devices is different. If the platform cannot change, the controls around it must: network isolation, restricted browsing or none, application allow-listing where the maker permits it, monitoring, and a replacement date set at purchase. The March 2019 chain started in a browser, and a device that cannot browse cannot be reached that way.
Mitigations get bypassed in time
A fifth objection is that exploit mitigations are not permanent. Attackers study them and find ways around them, so a newer platform’s advantage may fade. That may be so, and nothing in the March 2019 records says otherwise. But a mitigation that is eventually bypassed still raises the cost of attack while it holds, and the newer platform keeps receiving new mitigations as long as it is supported. The older platform, once its development has stopped, receives none. The gap between them may narrow for a given class of bug and widen for the next.
A platform register next to the patch report
Pulling the parts together, we think an organization’s monthly vulnerability report should carry a second table beside patch compliance. For each operating system version in use, it would show the number of systems by category (general desktop, application-bound, embedded, server), the vendor’s end of support date, whether extended updates are being bought, the owner of each category, and the date each group is due to move or be retired.
The register would answer the questions that March 7, 2019 posed. Which machines run the platform an exploited, unpatched flaw affects? Who uses them for web browsing? Who owns each group, and what can each owner do this week? In our reading, an organization with such a register could have acted on Google’s disclosure the same day. One without it had little to act on before March 12.
Who signs for an old platform
A register is useful only if someone acts on it. We would attach a decision to every group of systems on an older platform: a named owner, a reason the group has not moved, the compensating controls in place, and an expiry date for the decision. When the date passes, the owner either moves the systems or renews the decision with a fresh justification. That mirrors how organizations already handle exceptions to security policy, and it keeps old platforms from becoming permanent by default.
The register also gives procurement a role. When a new application, device or service is bought, the contract can state which operating system versions the vendor will support and for how long, and what happens when the vendor’s supported platform reaches its end of support before the contract ends. Those terms decide, years in advance, whether the application-bound and embedded categories will grow or shrink. An organization that writes them into its contracts is building its next migration plan at the point of purchase.
Starting small
An organization does not need a complete register to begin. It can start with the platform versions its endpoint tools already report, add the vendor’s end of support date for each, and assign an owner to every version that is not the current release. That first pass may take days rather than months. The second pass adds categories, separating ordinary desktops from application-bound machines, embedded equipment and servers. The third adds the systems the endpoint tools cannot see, from network discovery and the asset lists held outside IT.
Each pass makes the register more useful for the next disclosure like Google’s. After the first, the organization can say how many machines run an affected version. After the second, it can say who owns them and whether they can be restricted. After the third, it can say with more confidence that it has not missed a group. We would publish the register internally after the first pass, with its gaps marked, rather than wait for completeness. A visible, partial register invites the owners of unlisted systems to come forward; an unpublished, complete one does not exist.
Sector reporting
Banks, health systems and agencies each answer to different examiners and auditors, and we do not suggest a single format. But each of them reports on vulnerability management in some form, and each can add platform currency to what it already reports. For a bank, that may sit beside patch metrics in the information security report to the board. For a health system, it may sit with the medical device inventory, where embedded platforms are already tracked. For an agency, it may sit with the asset inventory it already keeps. The point is the pairing: wherever patch compliance is reported, platform currency should be reported beside it.
The register also gives boards and examiners a better picture than patch compliance alone. A high compliance rate on an estate full of old platforms may look like good hygiene and still carry exposure that patches cannot remove. A modest compliance rate on a current estate may carry less. Neither figure tells the story alone. Together, they show both the bugs that have been fixed and the classes of bugs that the platform itself makes harder to use.
With hindsight
The deadline the bridge was built for arrived on schedule. Microsoft’s lifecycle page lists the end of extended support for Windows 7 in January 2020, and the end of the third and last year of its Extended Security Updates in January 2023.
The 2019 flaws also came back as a compliance item for federal agencies. On November 3, 2021, CISA issued Binding Operational Directive 22-01, which established a catalog of known exploited vulnerabilities and required federal civilian agencies to remediate them on its timelines: within six months for vulnerabilities with a CVE ID assigned before 2021, and within two weeks for others, unless CISA adjusted the timeline for grave risk. The directive applied to all software and hardware on federal information systems, including systems operated by third parties on an agency’s behalf. CISA’s catalog shows both CVE-2019-0808 and CVE-2019-0797 added on November 3, 2021, with a due date of May 3, 2022. It shows the Chrome flaw from the same chain, CVE-2019-5786, added on May 23, 2022, with a due date of June 13, 2022. For agencies, the 2019 chain thus became a set of dated obligations, and any agency still running an affected system had to remediate it by those dates. CISA’s page for BOD 22-01 now says the directive has been revoked and superseded by BOD 26-04; the catalog itself remains in use.
The same pattern then repeated for Windows 10. Microsoft’s lifecycle page lists the retirement of Windows 10 Enterprise and Education in October 2025, and its Extended Security Updates FAQ lists Windows 10 with an end of support date of October 14, 2025. The FAQ says the editions that qualify for Windows 10 extended updates include Enterprise, Education, and Pro in commercial use, and describes the program in the same terms as before: a last resort paid option and a temporary bridge, offering critical and important updates for up to three years. An organization that kept a platform register through the Windows 7 transition had a ready-made plan for the Windows 10 one: the same categories, the same owners, the same question about which year of the bridge would be the last.
Microsoft’s FAQ also sets out the bridge’s dates. It lists the end of the first year of Windows 10 extended updates as October 13, 2026, the second as October 12, 2027, and the third as October 10, 2028. By this point in 2026, an organization buying that bridge has passed its first renewal point. The FAQ notes, too, that extended updates for select embedded products are available through the equipment makers rather than through volume licensing, which is a reminder that the embedded category has its own path and its own owner.
CISA’s guidance on the catalog speaks to the same record. On its directive page, CISA says it strongly recommends that organizations have updated asset inventory information to determine the software and hardware products in their environment and to identify the catalog vulnerabilities that directly affect them. It also says the catalog should be an input to an organization’s prioritization framework rather than the only criterion.
What held up is the distinction between patch currency and platform currency. The March 2019 chain showed it on a small scale: one exploited flaw that, on the reporter’s belief and the vendor’s product list, reached only the old platform, and another that the vendor rated as unlikely to be exploited on the latest release. What we would add today is that the known exploited vulnerabilities catalog made old flaws a dated obligation for federal agencies under BOD 22-01, and gave every other sector a public, dated list to measure itself against. A platform register tells an organization where those old flaws could still live.
Spotted an error? Tell us and we will check it and correct the article where needed.