Struts S2-057 was a test of knowing where the framework ran, and how

The Struts flaw of August 22, 2018 had a fix on day one. Using it required four records: where each copy ran, how it was configured, which fix it could take and when its branch would end.

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

The Struts flaw of August 22, 2018 had a fix on day one. Using it required four records: where each copy ran, how it was configured, which fix it could take and when its branch would end.

  1. The week On Wednesday, August 22, 2018, the National Vulnerability Database published CVE-2018-11776, a flaw in Apache Struts, the open-source framework that organizations use to build Java web applications.
  2. Location: the copies you built and the copies you bought The easy half of the location question is the code an organization’s own developers wrote.
  3. State: the settings that decided whether a copy was exposed S2-057 described the flaw as possible remote code execution when alwaysSelectFullNamespace was true, whether set by the user or by a plugin like the Convention Plugin, and then one of two patterns held.
  4. Change: an upgrade, a workaround or a vendor patch The bulletin’s recommendation was to upgrade to Struts 2.3.35 or Struts 2.5.17.
  5. Time: the branch clock the bulletin did not mention The fourth part of the inventory is the one that August’s bulletin did not raise.
  6. Two Struts flaws, two different questions Comparing the 2018 flaw with the one behind the Equifax breach shows why the state record matters.
  7. What each record holds Separating the four parts is useful only if each produces a record someone can keep.
  8. Running the response as a sequence When a bulletin like S2-057 arrives, the four records turn into a sequence of steps.
  9. Who owns each part The four parts belong to different people, and that is the practical point of separating them.
  10. Objections: scanners, vendors and the size of the job One objection is that a good vulnerability scanner makes the inventory unnecessary.
  11. How it lands in banks, health systems and agencies A bank that runs customer-facing Java web applications also buys platforms for contact centers and communications, and the Cisco list shows that such platforms can carry the same framework inside them.
  12. What we would ask a technology leader today In our reading, the useful questions are short and specific to the four parts.
  13. With hindsight The GAO report on the 2017 breach, dated August 30, 2018, filled in the history that made the August flaw important.

The week

On Wednesday, August 22, 2018, the National Vulnerability Database published CVE-2018-11776, a flaw in Apache Struts, the open-source framework that organizations use to build Java web applications. The record described the flaw as possible remote code execution in Struts versions 2.3 to 2.3.34 and 2.5 to 2.5.16. Remote code execution means that an outside party who can reach the application can make the server run commands of the attacker’s choosing.

The record did not describe a flaw that every Struts application had. It described a set of conditions. The first was a configuration setting called alwaysSelectFullNamespace being true, whether a developer had set it or a plugin, such as the Convention Plugin, had set it. The second concerned how the application’s configuration was written. Exposure arose when results were used with no namespace while the package above them had no namespace or a wildcard namespace. A similar exposure arose when a url tag in a page had no value and no action set, again under a package with no namespace or a wildcard namespace.

That same Wednesday, the Apache Struts project released Struts 2.5.17. The project’s releases page lists that version with the date 22 August 2018. The version before it on the 2.5 line, 2.5.16, carries the date 16 March 2018, and the releases page today lists S2-057, the project’s bulletin number for this flaw, among the vulnerabilities that apply to 2.5.16.

On Thursday, August 23, Cisco first published a security advisory on the same flaw as it applied to Cisco’s own products, which used the Struts library inside them. Its summary, in the final version last updated on September 17, 2018, described the cause in the same terms as the public record: results with no namespace value and url tags with no value or action, in cases where the configuration above them also had no namespace or a wildcard namespace. It said that an attacker could exploit the flaw by sending a request with malicious input to an affected application, and that success would let the attacker run code with the privileges of that application. Cisco went on updating the advisory after that week.

Two features of that week matter for the argument that follows. The fix shipped on the same day the vulnerability record appeared, so the question for any organization was not whether a patch existed. And a major vendor needed its own advisory to tell customers which of its products carried the framework. A security team that kept a list of applications its developers had built with Struts would not have found the copies inside purchased products by reading that list.

The flaw also arrived with history. In 2017, a different Struts vulnerability had been the way into Equifax, a fact the Government Accountability Office set out in a report dated August 30, 2018. We draw on that report in the analysis below. The 2018 flaw was a new test of the question the 2017 breach had raised: whether an organization knew, at the moment a framework flaw was announced, every place that framework was running.

Patching a well-known framework sounds like the easy part of security. A project announces a flaw, ships a fixed version and tells users to upgrade. CVE-2018-11776, the Apache Struts flaw published on August 22, 2018, had all of that on the first day. What it tested was everything an organization needed to know before an upgrade could start.

Our argument is that the Struts flaw of August 2018 was an inventory test with four parts, and that each part had a different owner. First, where the framework ran: in applications the organization built, and inside products it bought. Second, how each copy was configured, because the flaw needed particular settings to be exploitable. Third, which fix each copy could take, since an upgrade, a configuration workaround and a vendor patch were different jobs with different risks. Fourth, how long each copy’s branch would be supported, because the project would soon announce the end of one of the two branches it named. The parts do not overlap: the first is about location, the second about state, the third about change and the fourth about time. Together they cover what a leader needed to answer about any copy of the framework.

We take each part in turn, then test the argument against the strongest objections and look at how it lands differently in banks, health systems and public agencies. Organizations with legal obligations tied to vulnerability management should take advice from counsel on how those obligations apply to them. Nothing here describes what any particular law requires.

Location: the copies you built and the copies you bought

Applications built in house

The easy half of the location question is the code an organization’s own developers wrote. Struts is a framework, which means the application is built on it: the developers write actions, results and pages that Struts wires together. A team that knows it built an application on Struts can find it by searching the build files where the dependency is declared. That is still a search, not a lookup, unless someone maintains the answer before a flaw is announced.

The Equifax record shows what goes wrong when the answer is not maintained. According to the GAO report on the 2017 breach, Equifax officials said the Struts vulnerability of that year was not properly identified as being present on the company’s online dispute portal when patches were being installed across the company. After receiving a notice of the vulnerability from the United States Computer Emergency Readiness Team in March 2017, the officials said, they circulated it among their systems administrators, but the recipient list was out of date and the notice did not reach the individuals who would have been responsible for installing the patch. A network scan the company ran afterward, they said, did not detect the vulnerability on the portal.

Two separate failures sit in that account. One is the distribution list: the people who owned the application were not on it. The other is the scan: the tool that was supposed to catch what the list missed did not find the vulnerable copy. Neither failure is about patching. Both are about knowing where the framework ran and who answered for it.

Products that carried the framework inside

The harder half is software the organization bought. Cisco’s advisory on CVE-2018-11776, first published on August 23, 2018 and last updated on September 17, 2018, lists twenty products and services under its vulnerable products section. They span contact center software, voice and unified communications, collaboration, video distribution, network security and a cloud-hosted analytics service. None of them is a Struts application in the sense a buyer would recognize. Each is a product that happens to contain the library.

A security team that relied on its own developers’ dependency lists would not have seen any of them. Neither would a team that relied on the vulnerability record alone, because the record named Apache Struts, not the products built on it. The information arrived from the vendor, in the vendor’s format, on the vendor’s schedule, and it kept changing: the advisory was updated after its first publication, and the fixed release column lists patch files and releases dated from late August through late September 2018.

This is the part of the inventory that the buyer cannot build alone. The buyer can list its products and versions. Only the vendor can say which of them contain the library, at which version, and whether the product uses it in a way that exposes the flaw. The practical consequence is contractual and operational at once: the vendor’s duty to tell customers what is inside a product, and the customer’s process for turning that information into a ticket against a named system with a named owner.

Records that do not agree

Even the reference records described the affected range differently. The vulnerability database entry said Struts versions 2.3 to 2.3.34 and 2.5 to 2.5.16. The project’s bulletin, S2-057, listed the affected software as Struts 2.0.4 through 2.3.34 and Struts 2.5.0 through 2.5.16. An inventory tool that matched installed versions against the database’s range would have treated an application on an older 2.x release differently from one that matched against the bulletin.

The gap matters for old applications, built on early versions and never upgraded, which are the ones an inventory is likeliest to have lost track of. In our view, an organization should match its inventory against the project’s own bulletin as well as the vulnerability database, and treat any disagreement between them as a reason to look, not as a reason to assume safety.

State: the settings that decided whether a copy was exposed

A flaw that needed conditions

S2-057 described the flaw as possible remote code execution when alwaysSelectFullNamespace was true, whether set by the user or by a plugin like the Convention Plugin, and then one of two patterns held. Results were used with no namespace while their upper package had no namespace or a wildcard namespace. Or a url tag had no value and no action set while its upper package had no namespace or a wildcard namespace. The bulletin gave the flaw a maximum security rating of Critical and said that all Struts 2 developers and users should read it.

Conditions like these cut two ways. They meant that some applications on affected versions were not exploitable through this path. They also meant that proving an application was not exploitable required reading its configuration, its plugins and its page templates, and keeping that answer current as developers changed them. A setting introduced by a plugin is the hardest case, because the developer who added the plugin may not have known it changed how namespaces were selected.

Contains the library is not the same as exposed

Cisco’s advisory made the distinction explicit. Products marked with an asterisk, it said, contain an affected Struts library, but due to how the library is used within the product, they are not vulnerable to any of the exploitation vectors known to Cisco at the time of publication. Eighteen of the twenty entries carry the asterisk. The two that do not are the Identity Services Engine and the cloud-hosted Network Performance Analysis service.

That is a useful piece of vendor analysis, and it carries two qualifications a buyer should not lose. The judgment rests on vectors known to the vendor at publication, so it can change. And every asterisked product was still listed with a bug identifier, and all but one with a fixed release, which suggests, in our reading, that the vendor meant to update the affected library anyway. For an asset owner, the asterisk changes the urgency of a fix, not whether a fix belongs on the plan.

Who records the configuration

In our judgment, configuration is the part of the inventory that a scanner is least able to read from outside. A version can be read from a file name or a manifest. Whether alwaysSelectFullNamespace is true, and whether every package and result has a namespace, is a property of how the application was written. The developer team is the only party that can answer it with confidence for in-house code, and the vendor is the only party that can answer it for a product.

We think the answer belongs in the same record as the version: for each copy of a widely used framework, the version, the plugins in use and the security-relevant settings the project has warned about. When a bulletin names a setting, the organization can then search its own records rather than its source code. When the record is missing, every new bulletin starts a code review under time pressure.

Change: an upgrade, a workaround or a vendor patch

The bulletin’s recommendation was to upgrade to Struts 2.3.35 or Struts 2.5.17. On backward compatibility it said that both versions contained the security fixes only, and that no backward incompatibility issues were expected. It then added that the project had received reports that compatibility issues could occur in some cases, related to the direct use of ArrayList in conversion logic, that a warning would appear in the logs, and that a constant could be defined in the application’s configuration to address it. It said the project was working on a new release to fix that problem.

That paragraph is a small case study in why upgrades stall. A release described as security fixes only still changed behavior for some applications. The fix for the side effect was another configuration change. An organization that wanted to move quickly needed a way to test each application, read its logs for the specific warning and apply the constant where needed, which again depended on knowing the application and having a test path for it.

The workaround the project called weak

The bulletin also gave a workaround and described it, in its own words, as a temporal weak workaround. It told users to verify that they had set a namespace for all defined packages, or for all defined results where applicable, and that they had set a value or action for all url tags in their pages when the upper package had no namespace or a wildcard namespace. It urged users to upgrade as soon as possible because the new versions also contained what it called critical overall proactive security improvements.

A workaround of that kind is a configuration audit, not a patch. It can be the right temporary step for an application that cannot be upgraded in the time available. It also leaves behind a condition that must stay true: the next developer to add a package or a result has to keep setting the namespace. If the workaround is recorded only in a ticket that was closed, the protection decays without anyone noticing.

The patch only the vendor could ship

For bought products, neither the upgrade nor the workaround was available to the customer. The customer could not replace a library inside a vendor’s product without breaking its support, and could not edit the product’s configuration files in the way the workaround required. The fix was whatever the vendor shipped and when. Cisco’s fixed release column shows the range: patch files for some products, engineering special releases or standalone packages for others, and, for the video distribution product, Struts 2.3.35 itself, listed with a date of 15 September 2018.

The customer’s job in that case was to track a vendor’s schedule for each affected product, apply each patch through the vendor’s process, and decide what to do in the meantime about exposure. For a product reachable from the internet, the interim choice is between restricting access and accepting the risk until the patch lands. That choice belongs to the business owner of the product, informed by the security team, and it should be written down with a date.

Time: the branch clock the bulletin did not mention

The fourth part of the inventory is the one that August’s bulletin did not raise. On November 14, 2018, the Struts project announced that the 2.3.x line would reach its end of life in six months, and that from that date it would support 2.3.x only in case of security vulnerabilities. The announcement said that some security-related changes could not happen without architectural changes that would affect backward compatibility, which was what had happened on the 2.5.x line. It said there was no urgent issue forcing the immediate removal of 2.3.x, but that users should consider migrating, and it advised against starting new projects on 2.3.x.

The project’s end-of-life page now records the dates. Struts 2.3.x reached end of life on 12 September 2019. Struts 2.5.x reached end of life on 30 October 2023. Struts 1.x had reached end of life on 5 April 2013. The page says that when a Struts version reaches end of life, the project no longer provides security patches, bug fixes or updates for that branch, and that end-of-life versions receive no support at all from the project.

An application upgraded in August 2018 to Struts 2.3.35 had done what the bulletin asked. It was also on a branch that, by the date the project later recorded, would reach end of life on 12 September 2019. An inventory that recorded only whether each copy was patched against the latest bulletin would have shown that application as compliant. An inventory that also recorded the branch and its support end date would have shown a migration that needed funding.

The branch clock bites hardest on the products and applications that are hardest to change. Those are the ones where a team accepts a security-only upgrade because a branch migration is too large, and then finds that the branch itself has a date. Planning for that date is a budget and staffing question, which is why we treat it as a separate part of the inventory with its own owner.

Two Struts flaws, two different questions

Comparing the 2018 flaw with the one behind the Equifax breach shows why the state record matters. The Known Exploited Vulnerabilities catalog describes the 2017 flaw, CVE-2017-5638, as one in which the Jakarta Multipart parser in Struts allows malicious file upload using the Content-Type value, leading to remote code execution. That description names no configuration conditions. For a flaw of that kind, the location and version records were close to enough: if the affected version and parser were present, the copy needed fixing.

The 2018 flaw was different. Its exposure depended on a setting that a plugin could change and on how results and url tags were written. For a flaw of that kind, location and version produce a list of candidates, not a list of exposed systems. The state record is what turns candidates into findings. An organization that had responded to 2017 by building a good location record had done the necessary work for 2017 and half the work for 2018.

The comparison also shows that the four parts carry different weight from one flaw to the next. A flaw with no conditions puts the weight on location and change. A flaw with conditions puts it on state. A flaw announced near the end of a branch puts it on time. Because nobody can know in advance which kind the next bulletin will be, all four records have to exist before it arrives.

There is a further lesson in how the 2018 flaw was described. The bulletin’s text about conditions was precise but dense, and it was repeated almost word for word in the vulnerability database and closely paraphrased in the Cisco summary. Turning that text into a check an owner could run against an application took engineering judgment. We think the security team’s job after a bulletin like S2-057 includes translating its conditions into a short checklist each owner can apply, and recording that checklist with the response, so the reasoning behind each finding of not exposed can be reviewed later.

What each record holds

Separating the four parts is useful only if each produces a record someone can keep. We describe what we think each record needs to contain. These are our recommendations, not requirements drawn from any rule.

The location record lists every system where the framework runs. For in-house applications, that means the application name, the repository, the build file that declares the dependency and the environments where it is deployed. For bought products, it means the product name and version as the vendor names them, the hosts or services where the product runs, and the contract that governs the vendor’s duty to disclose what is inside. The test of the record is whether a vendor advisory listing product names, like Cisco’s, can be turned into a list of installed systems without a meeting.

The state record holds, for each copy, the framework version and the settings the project has warned about in past bulletins. For Struts after August 2018, that would include whether alwaysSelectFullNamespace is true, which plugins are in use, and whether every package and result declares a namespace. For bought products, the state record holds the vendor’s own statement, such as Cisco’s asterisk, with the date it was made and the qualification attached to it.

The change record holds, for each copy, which remedy applies and its status: an upgrade with its test results, a workaround with the condition it depends on, or a vendor patch with the vendor’s date. It also holds the interim decision about exposure, who made it and when it will be reviewed. The workaround entry deserves particular care in audit, because a workaround is a promise about future configuration that someone must keep.

The time record holds, for each copy, the branch and the project’s published end-of-support date, or the vendor’s support date for the product version. It feeds the budget. A branch whose end-of-life date falls inside the planning period should show up as a funded migration or as an accepted risk signed by someone with the authority to accept it.

Running the response as a sequence

When a bulletin like S2-057 arrives, the four records turn into a sequence of steps. The order matters, because each step depends on the one before it.

The sequence starts with location. The security team matches the bulletin’s affected versions, and the vulnerability database’s, against the location record and against vendor advisories as they arrive. The output is a list of systems with owners, split into in-house copies and bought products. Every owner on the list receives the bulletin directly. The Equifax account shows why the distribution has to come from the record rather than from a mailing list maintained elsewhere: the people who needed the notice were not on the list that received it.

The second step is state. Each owner checks the conditions the bulletin names against the state record, and records a finding: exposed, not exposed with a reason, or unknown. Unknown is treated as exposed. For bought products, the finding is the vendor’s statement until the vendor changes it.

The third step is change. For exposed and unknown copies, the owner chooses a remedy and records it, with interim restrictions where the remedy will take time. For internet-facing systems, the interim decision is made quickly and written down. For the workaround, the condition it depends on is added to the state record so that the next code change is checked against it.

The fourth step is time. Every copy touched in the response has its branch checked against the end-of-support dates. A copy that was patched on a branch nearing its end is added to the migration plan. This step is easy to skip under pressure, and it is the one that prevents the next bulletin from finding the same copies on a branch the project no longer fixes.

Each step leaves evidence. Taken together, the evidence shows what the organization knew, when it knew it and what it decided. That is the record an examiner, an auditor, an insurer or a board will ask for after an incident, and it is far easier to keep as the response happens than to reconstruct later.

Who owns each part

The four parts belong to different people, and that is the practical point of separating them.

  • Location belongs to whoever owns the software inventory, which spans application development, procurement and vendor management. The developers can list what they built. Only procurement and vendor management can make vendors say what is inside what they sold.
  • State belongs to the application owner and the developers or vendor behind each copy. It is the record of settings and plugins that decides whether a bulletin applies.
  • Change belongs to the operations and release teams that can test and deploy an upgrade, apply a workaround safely or install a vendor patch, and to the business owner who decides on interim exposure.
  • Time belongs to the people who plan and fund technology, because a branch reaching end of life is a migration project, not a patch.

The Equifax account in the GAO report touched three of the four. The notice did not reach the right administrators, which is location and ownership. The scan did not find the vulnerable copy, which is location again. The patch was not applied to the portal, which is change. The fourth part did not arise there, but it arose for every organization that answered the 2018 bulletin with a 2.3.35 upgrade.

Objections: scanners, vendors and the size of the job

A scanner would have found it

One objection is that a good vulnerability scanner makes the inventory unnecessary. The Equifax record does not support that. According to the GAO report, Equifax officials said that a scan run after the 2017 vulnerability was announced did not detect it on the portal. A scanner is one input to the inventory. It is not the inventory, and we doubt it can reliably read configuration conditions of the kind S2-057 described, or see inside a vendor’s appliance the way the vendor can.

Vendors are responsible for their products

A second objection is that the vendor, not the customer, owns the risk in a bought product. Contractually that may be partly true. Operationally the customer still runs the product, decides whether it faces the internet and lives with the consequences of a breach through it. Cisco’s advisory shows a vendor doing the analysis well, and it still left each customer to map product names to installed systems, schedule patches and decide on interim exposure.

The job is too large to keep current

A third objection is that recording version, settings and support dates for every copy of every framework is more than any team can maintain. We agree that it cannot be done for every library at the same depth. It can be done for the frameworks whose flaws reach remote code execution on internet-facing systems, a list each organization can keep short by its own choices. Struts was on that list in 2017 and again in 2018. A short, accurate list for those frameworks is worth more than a long, stale list for everything.

A software bill of materials settles it

A fourth objection is that a machine-readable list of components shipped with each product, a software bill of materials, answers the location question for bought software. It can answer part of it, if the vendor provides one and keeps it current. It does not answer the state question, because a list of components does not say how they are configured or whether the product uses the affected code path, which is exactly what Cisco’s asterisk addressed. It does not answer the change or time questions either. A component list is a good input to the location record, and it is only that.

The flaw needed special conditions, so the risk was small

A fifth objection is that the conditions narrowed the risk enough to wait. That may have been true for a given application. The difficulty is that nobody could know it without the configuration record, and the conditions could be introduced by a plugin. The federal government later listed this flaw as one known to have been exploited, as we discuss in the hindsight section. Waiting was a defensible choice only for an organization that had checked.

How it lands in banks, health systems and agencies

Banks and other financial firms

A bank that runs customer-facing Java web applications also buys platforms for contact centers and communications, and the Cisco list shows that such platforms can carry the same framework inside them. The contact center and the phone system are software estates with their own libraries. For a bank, the location question has to cover those platforms as well as online banking, and the vendor management program is the natural place to require timely disclosure of what is inside them.

Health systems

A health system whose estate is weighted toward purchased clinical and operational systems, which it cannot modify, will find that the change part of the inventory consists largely of vendor patches, and the interim decision about exposure is the one they control. A health system that knows which purchased systems face the internet can decide quickly which ones to restrict while it waits for a vendor.

Public agencies

Agencies face the same mix. Where an application was built under a contract that has since ended, the state part of the inventory is hard to reconstruct, because the people who knew the configuration have gone. The Equifax report also shows agencies on the other side of the problem, as customers of a vendor that was breached. According to GAO, the Internal Revenue Service, the Social Security Administration and the U.S. Postal Service, three of Equifax’s major federal customers for identity verification, assessed the company’s controls and adjusted their contracts after the breach, and the IRS terminated one of its contracts. An agency’s inventory question therefore extends to the services its vendors run on its behalf.

GAO also recorded what the central agencies did. Equifax did not ask the Department of Homeland Security to help respond, GAO said, but the department reminded federal agencies of the importance of correcting the software vulnerability that led to the breach. A reminder of that kind reaches agencies through their chief information officers. Whether it reaches the people who run each affected system depends on the same location record we have described. For agencies, the reminder from the center and the record at the edge are two halves of one process, and the 2018 flaw tested both.

The customer side of the Equifax story deserves its own entry in an agency’s records. GAO reported that the agencies that used the company’s services conducted independent assessments of its security controls, revised their own identity proofing processes and changed their contracts. Each of those steps is easier when the agency already knows which vendor services handle its data and what the contract lets it ask for. An agency’s location record should therefore include services run on its behalf, and the contract terms that let it ask a vendor where a framework runs inside them.

What we would ask a technology leader today

In our reading, the useful questions are short and specific to the four parts.

  • For each framework on our short list, can we produce, within a day, every system that runs it, including purchased products, with an owner for each?
  • For each copy, do we record the version, the plugins in use and the settings the project has warned about?
  • When a fix is a vendor patch, who decides on interim exposure, and where is that decision recorded with a date?
  • Do we know the end-of-support date for each branch we run, and is there a funded plan for the ones that end within our planning horizon?

These questions do not require new tools. They require that the answers be kept before the next bulletin, by the people who own each part.

With hindsight

The GAO report on the 2017 breach, dated August 30, 2018, filled in the history that made the August flaw important. It said that on March 10, 2017, unidentified individuals scanned Equifax’s systems for a Struts vulnerability that the United States Computer Emergency Readiness Team had publicly identified, in GAO’s words, just 2 days earlier, and found a server housing the online dispute portal that ran a vulnerable version. Beginning on May 13, 2017, attackers gained access to the portal and extracted data, and the company discovered the breach on July 29, 2017. GAO reported that the attackers accessed personal information of at least 145.5 million individuals, and that Equifax identified four major factors that allowed the breach: identification, detection, segmentation of access to databases and data governance. On detection, an expired digital certificate had stopped a tool from inspecting network traffic; GAO said the certificate had expired about 10 months before the breach occurred.

The federal government later recorded both Struts flaws as exploited. The Cybersecurity and Infrastructure Security Agency’s Known Exploited Vulnerabilities catalog, in the version dated November 13, 2026, lists CVE-2018-11776 as an Apache Struts remote code execution vulnerability, added on November 3, 2021, with a required action to apply updates per vendor instructions and a due date of May 3, 2022. It lists the 2017 flaw, CVE-2017-5638, with the same added date and marks its use in ransomware campaigns as known. For the 2018 flaw, that field reads unknown.

The catalog has kept adding Struts entries. On October 8, 2026, it added CVE-2016-3081, a Struts command injection flaw, with a due date of October 11, 2026. Its required action refers to a CISA directive, BOD 26-04, titled Prioritizing Security Updates Based on Risk, and to forensics triage requirements, and it tells stakeholders to evaluate each asset’s internet exposure and to discontinue use of the product if mitigations are unavailable. A flaw from 2016 was still being added to a federal list of exploited vulnerabilities in October 2026, which says something about how long old framework versions stay in service.

The branch clock ran later than the November 2018 announcement suggested, but it ran. That announcement had given the 2.3.x line six months; the project’s end-of-life page records 12 September 2019 as the end-of-life date for Struts 2.3.x and 30 October 2023 for Struts 2.5.x. The project’s releases page today lists Struts 7.4.0 as the best available release and lists 2.5.17, the August 2018 fix, among prior releases that it says were reclassified as not recommended because they contain security issues. The end-of-life page now also points to third-party extended support for end-of-life versions, which the Apache Software Foundation says it does not endorse.

What held up is the shape of the problem. In our reading, each later Struts entry posed the same four questions: where the framework runs, how it is configured, which fix applies and how long the branch lasts. What did not hold up is any idea that an August 2018 upgrade settled the matter. The fixed versions of that week became versions to replace. The organizations that kept a live record of where the framework ran, and when each branch ended, could answer the next bulletin from their records. The rest had to start again with a search.

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

  • Apache Struts
  • CVE-2018-11776
  • software inventory
  • vulnerability management
  • end of life
  • Equifax