Deception CheckResearch & field notes
DETECTION ENGINEERING / REMOTE ACCESS

Ivanti exploitation: why a clean integrity check is not enough

October 14, 2026 · Nick Carroll, Co-Founder & CTO

The patch addresses the entry point. It does not tell you whether someone already used it, which credentials they reached, or what happened behind the gateway.

Volexity's investigation of active Ivanti Connect Secure exploitation describes a difficult response problem: attackers could run commands through the exploit channel, operate in memory, and remove selected local evidence. A persistent web shell was not required. A clean appliance check is useful evidence, but it is not an all-clear.

This is worth acting on now. Ivanti released fixes on October 13, and CISA added CVE-2026-104935 to its Known Exploited Vulnerabilities catalog that day. The October 16 remediation deadline applies to U.S. federal civilian executive branch agencies under the KEV program. Other operators should treat the confirmed exploitation as a reason to act promptly, not as permission to wait for that date. Ivanti's update; CISA's KEV catalog.

RESPONSE PRIORITY

Close the entry point. Keep the investigation open.

Apply the vendor's fix for your supported release branch. Coordinate containment and evidence collection before disruptive recovery work. Review downstream identities and systems, not only the gateway's filesystem.

What the investigation actually found

Volexity describes three flaws in a chain. The authentication bypass and file-write/code-execution issues are covered together by CVE-2026-104935; a separate privilege-escalation flaw completed the path to root. The researchers observed repeated exploitation beginning October 3 against an appliance that was fully patched at that time. That is different from an operator ignoring an available fix.

In the investigated attack, Volexity identified no first-stage backdoor or web shell. The actor repeatedly reused the exploit channel for commands and results, executed Python in memory, and cleaned selected debug and audit records. Its investigation also found credential collection and Active Directory queries, including searches for LAPS data and BitLocker recovery information. Those findings come from Volexity's incident investigation, not our research fleet.

WHY ONE APPLIANCE CHECK CANNOT ANSWER THE WHOLE QUESTION
01 / ENTRY

Repeated HTTPS exploitation

Commands and results use the exploit channel. A separate command-and-control connection is not required.

02 / EXECUTION

Transient files, memory, and cleanup

Some evidence may be brief or absent from a later filesystem check.

03 / REACH

Credentials and directory access

Identity logs and downstream records can answer questions the appliance alone cannot.

Attack behavior summarized from Volexity. The evidence checks below are our defensive application of those findings; this is not a replay of a Deception Check capture.

Look in three places, not one

The practical question is whether your evidence covers the access path, the execution, and the systems reached afterward. A negative result in one of those places does not fill a gap in the other two.

At the access path

Retain reverse-proxy, firewall, flow, and available HTTP records outside the appliance. Look for unusual request volume and repeated access to welcome.cgi. Where headers are captured, inspect inconsistencies between the requested path and X-DSCGI-PATH.

These are investigation leads, not proof by themselves. Many logging configurations do not retain the relevant headers or decrypted content.

On the appliance

Preserve available logs and arrange memory and filesystem collection with the response team. Know what your integrity tool checks and when its results were collected. Rebooting can destroy volatile evidence; a reset can destroy more.

Do not leave an exposed gateway reachable just to collect more evidence. Coordinate collection with containment and operational needs.

Behind the gateway

Review directory service accounts, LDAP activity, authentication records, password-spray indicators, and subsequent access to sensitive systems. Establish what a compromised gateway identity was allowed to read or do.

For a confirmed compromise, include stored and accessible secrets in recovery planning. A firmware update does not revoke a stolen credential.

The request-pattern leads above come from Volexity. The collection and recovery priorities also align with CERT-FR's October 14 alert. Keep source events and collection times alongside the analyst's conclusion so a later discovery can change the assessment without rewriting the original record.

Which fix, and what is still pending?

Ivanti assigns CVSS 8.6 to CVE-2026-104935. Its October patch summary lists multiple resolved versions because the product has separate release branches. This is not a single version threshold across every installation.

Published fixes for CVE-2026-104935, checked October 14
Installed branchPublished fixed release
22.822.8R2.5
25.1.125.1.1.2
25.1.2 / 25.1.3 upgrade path25.1.3.1; Ivanti also lists 25.1.3.0 as resolved

Confirm the supported upgrade path and current package against Ivanti's advisory before changing an appliance. CERT-FR's current alert identifies 25.1.3.1 as the fixed 25.1.3 release. Ivanti's CVE record says 22.7R2.14 and earlier are not affected by this particular flaw; that is not a recommendation to downgrade or a statement that those versions have no other vulnerabilities.

At the time of this review, the separate privilege-escalation fix was still described as forthcoming. Do not wait for that fix to close the already-patchable unauthenticated entry point. Equally, do not treat closing that entry point as proof that an earlier compromise has been removed.

Preserve, contain, and recover deliberately

  1. Identify the affected appliance and the access it provides. Record the installed build, exposure, service accounts, and routes into business or operational systems. Assign an owner for the response.
  2. Coordinate isolation and evidence collection. CERT-FR recommends isolation from Internet and internal networks and, where possible, memory and filesystem preservation before shutdown. Plan around safety and service continuity with the operational owner.
  3. Patch and investigate. Use the current vendor procedure and integrity tools, alongside the external evidence. Some collection or external integrity-check procedures can interrupt service; account for that in the change plan.
  4. If compromise is found, recover the trust boundary. Follow vendor and incident-response guidance for reset or rebuild, fixed firmware, credential and certificate replacement, and review of downstream access. Restore service with monitoring and an accountable decision, not just a green check.

These priorities follow CERT-FR's response guidance. The exact sequence depends on the appliance, evidence available, and operational constraints. Preserve what you can without postponing necessary containment.

Why this belongs in an OT conversation

Connect Secure is a remote-access gateway, not an industrial controller. The OT question is where its access leads. If a gateway identity can reach an engineering workstation, a historian, a vendor-support host, or a directory account used in operations, those paths belong in the investigation.

Our goal is to make unexpected use of those paths easier to notice and easier to investigate. An isolated decoy placed on a route that should not receive ordinary access can supply a distinct warning signal. Exporting that record off-host and verifying delivery to a named responder makes it more useful during recovery. That is a detection design to commission and test, not a substitute for patching, segmentation, or forensic collection.

We explore the evidence-retention side in The evidence that survives the wipe. The same commissioning question applies here: if the sensor is reset or taken offline, can the response team still retrieve the record of the interaction?

Make the evidence path part of the design

If your team is reviewing remote access into operational environments, we can help identify useful decoy placements and test whether an interaction produces retained evidence and reaches a responder.

Talk with Deception Check