The Ivanti VPN appliances broken into this October were current on their updates, and the vendor's own integrity check found nothing. The fix for part of the flaw had shipped in July, on a different release line. Most patch dashboards cannot see that gap.
On October 29, Ivanti shipped its second round of emergency fixes in a month for Connect Secure, its SSL VPN. The next day CISA added both newly patched flaws to its Known Exploited Vulnerabilities catalog and gave federal agencies until November 2 to fix the more serious one and to carry out forensic triage. Any organization that had dutifully applied the October 13 patches now had to upgrade again.
The sequence is familiar to anyone who runs edge appliances. What makes this one worth a CIO's time is the state of the first known victim. According to Volexity, which found the intrusion while investigating suspicious activity for a monitoring customer, the appliance was running 22.8R2.4, which Volexity described as fully patched at the time. An attacker it tracks as UTA0576 chained an authentication bypass, a file-upload race condition and a privilege escalation to get root on it, starting on October 3. A decoy network run by Defused saw attacks it attributes to the same entry flaw from October 1.
Patch compliance, a clean integrity scan and a moderate severity score are three of the most common ways a security program tells the board an appliance is fine. In this case all three were reporting accurately. They were measuring the wrong things.
A fix can exist and still not reach you
Ivanti's own account, given to Help Net Security on October 14, is the detail to dwell on. The company said it had found a portion of the vulnerability through its product security program and fixed it in release 25.1.3.0 in July. It "had been working to backport the fix to the remaining supported versions when exploitation occurred."
So for roughly three months, a fix for part of this chain sat in one release line while customers on other supported lines stayed exposed. Ivanti has not said the July release came with a public advisory, and the CVE for the entry flaw was published only on October 13. Those customers' patch dashboards were green because, measured against the line they were on, nothing was outstanding.
That is not a scandal peculiar to Ivanti. Every vendor that maintains several supported branches fixes the newest one first and works backward, and backporting into older code is slow, careful work. The trouble is that most enterprises have never asked how long the lag runs, or which branch receives fixes first. "Fully patched" is a statement about a version string. It says nothing about whether the vendor already knows of a flaw that your branch has not yet been fixed for.
Ivanti's current guidance makes the point for it: customers on 22.8 are told to move to 25.1.1.3 or later, with 22.8R2.6 as the fallback if migration cannot happen right away. The older line is still supported. It is plainly not where fixes land first.
A green patch dashboard tells you you've installed what the vendor shipped for your version. It cannot tell you what the vendor has already fixed somewhere else.
The integrity check graded its own homework
The second assurance failed more quietly. Ivanti supplies an Integrity Checker Tool, and many teams treat a clean run as the end of an investigation. Volexity found that UTA0576's chain patched library code only in memory, restored the configuration data it had tampered with, and deleted entries from the debug and audit logs. In Volexity's words, the integrity tool "does not detect any modifications." Reconstructing the attack took memory images, disk images and core dumps collected with Ivanti's help.
That has a budget consequence. If the vendor's check cannot clear an appliance, someone has to be able to capture memory from it before a reboot wipes the evidence. Ivanti's advice now includes preserving memory and logs before resetting where feasible. Few infrastructure teams have rehearsed doing that to a production VPN at two in the morning, and many incident-response retainers do not cover appliance memory forensics at all.
A 4.3 that ended in root
The third assurance is the severity score. One of the October 29 flaws, CVE-2026-118527, is rated 4.3 by Ivanti. On its own it lets an authenticated user upload files to a known location. Many enterprise patch policies give a score in that range weeks or months. Chained with the other two, it formed part of the route that gave an unauthenticated attacker root on the box. Rapid7's assessment was that the combined impact warrants emergency remediation outside normal patch cycles, whatever the individual scores say.
Figure 1
Assurance | What it actually measures | What October showed | Better question |
|---|---|---|---|
"Fully patched" | Installed release matches the latest one for your branch | Part of the flaw was fixed in another branch in July; the victim on 22.8R2.4 was current and exposed | Which branch gets fixes first, and how long does backporting take? |
Integrity check passes | Files and configuration on disk match the vendor's baseline | Changes lived in memory; configuration was restored and logs trimmed | Can we capture appliance memory before a reboot? |
Low CVSS score | Severity of one flaw in isolation | A 4.3 helped form a chain that ended in unauthenticated root | Is this flaw on an internet-facing device that has been chained before? |
MFA enforced at the VPN | The login flow requires a second factor | SonicWall's November 6 flaw can let a valid alternate credential bypass SAML-enforced MFA | Which login paths does the appliance accept besides the one we configured? |
Patched now | The known flaws are closed | October 13 patchers had to upgrade again on October 29 | Did anyone check for compromise before, and after, each patch? |
Five common assurances about an edge appliance, and what each one can and cannot tell you. Editorial assessment built from Volexity's October 13, 2026 analysis, Ivanti's statements as reported by Help Net Security and SecurityWeek, Rapid7's November 2, 2026 summary, and SonicWall's November 6, 2026 advisory as reported by SecurityWeek. Not a control framework.
What the attackers wanted was the keyring
Treat the VPN as a perimeter device and the incident reads as a perimeter breach. Look at what UTA0576 did once inside and it reads differently. Volexity documents extraction of the domain credentials the appliance stores for LDAP, password spraying, harvesting of credentials from captured traffic, Kerberoasting tools, and a custom LDAP tool that searched Active Directory for Windows LAPS local administrator passwords and BitLocker recovery keys using a compromised service account. Volexity assessed with high confidence that this was preparation for lateral movement. It was caught before the attackers' objective became clear.
The appliance was valuable because of what it was trusted to read. A remote-access gateway that authenticates against the directory carries directory credentials, and if the account it uses can read LAPS or BitLocker attributes, a compromised VPN hands over the recovery keys to every laptop in the estate. That is an identity-tier asset dressed as network hardware, and it is usually patched, monitored and budgeted as network hardware.
Nor is this an Ivanti-only story, which is the tempting conclusion. On November 6, SonicWall disclosed a flaw in its SonicOS SSL VPN, added to CISA's catalog the same day, that can let a user holding valid credentials for an alternate login method get around SAML authentication and the MFA enforced through it. SonicWall says it has investigated a report of potential exploitation; there is no workaround. CISA's catalog now lists 20 SonicWall-related vulnerabilities. Swapping vendors changes the logo on the next advisory.
Five questions for your next remote-access review
Branch. Which release line is each edge appliance on, and does the vendor fix that line first or backport to it later?
Lag. Will the vendor state, in writing, its typical backport window for security fixes across supported branches?
Evidence. Can your team or your retainer capture appliance memory and logs before a reboot, and has anyone practiced it?
Reach. What can the appliance's directory service account read? Can it see LAPS passwords or BitLocker recovery keys?
Severity. Does your patch policy treat any flaw on an internet-facing appliance with prior chained exploitation as urgent, regardless of score?
What this means for technology leaders
Report edge appliances to the board by branch and lag, not by compliance percentage. A VPN that is current on an older line deserves a different color from one on the line the vendor fixes first, and the plan to migrate it belongs in next year's budget rather than in the next incident review. Moving from 22.8 to 25.1 is a project with downtime and testing attached. It is cheaper to schedule than to be forced into.
Shrink what the box can read. Scoping the appliance's service account away from LAPS and BitLocker attributes is a modest piece of directory work, and it changes the worst case of the next VPN zero-day from estate-wide credential theft to a contained network incident.
Put the backport question into the renewal. Vendors are rarely asked to commit to how quickly a fix reaches every supported branch, which is one reason the answer is rarely good. A buyer that asks, and writes the answer into the support terms, at least learns how long its own exposure window can run.
And stop treating a clean integrity scan as closure. In October it was the attacker who decided what that scan would find.
Sources and notes. Volexity, "A Series of Unauthenticated Events: Chained Exploitation of Ivanti Connect Secure Zero-Days" (October 13, 2026, updated October 29, 2026): victim appliance on 22.8R2.4, described as fully patched at the time; exploitation from October 3; investigation from October 5; three-part chain; in-memory patching, configuration restoration and log deletion; Integrity Checker quote; credential theft, AD queries, LAPS and BitLocker searches, Kerberos tooling; high-confidence lateral-movement assessment. Zeljka Zorz, Help Net Security (October 14, 2026): Ivanti statement on the July 25.1.3.0 fix and backporting; Defused activity from October 1. Eduard Kovacs, SecurityWeek (November 2, 2026): October 29 fixes, fixed versions, requirement to upgrade again, Defused decoy observations. Zeljka Zorz, Help Net Security (November 2, 2026): CISA deadlines and forensic triage requirement; Ivanti advice on preserving memory and logs and rotating credentials. Rapid7, "Additional Ivanti Connect Secure zero-days exploited in the wild" (November 2, 2026): CVSS scores, affected and fixed versions, Ivanti's 22.8 to 25.1 migration guidance, emergency-remediation assessment, October 30 KEV addition. Eduard Kovacs, SecurityWeek (November 7, 2026): SonicWall advisory SNWLID-2026-0014, CVE-2026-19868, KEV addition November 6, count of SonicWall entries in KEV. Ivanti's advisory page could not be loaded directly; Ivanti statements are as reported by the outlets above. Figure 1 is editorial assessment. Journalism, not procurement or legal advice. Corrections welcome.



