Ivanti Connect Secure zero-day chain leaves nothing for the integrity checker to find
An unknown actor gained root on fully patched Ivanti VPN appliances without an implant, re-exploiting for every command. Two flaws are fixed; the privilege escalation step is not, and Ivanti's own checker does not see the intrusion.

A previously undocumented actor gained unauthenticated root on a fully patched Ivanti Connect Secure (ICS) VPN appliance and left nothing on disk for the vendor’s own integrity tooling to find. Volexity, which found the activity at a customer, says the actor chained three flaws, installed no web shell or backdoor, and re-ran the full exploit chain for every command it issued [1]. Ivanti fixed two of the three flaws on 13 October under CVE-2026-104935. The privilege escalation step has no fix yet [1][5]. For defenders, the practical problem is detection, not only patching.
What the advisory and the record say
The CVE record, published by Ivanti as CNA on 13 October, describes an authentication bypass by spoofing (CWE-290) in ICS before 25.1.3.0 that lets a remote unauthenticated attacker execute arbitrary commands. It scores 8.6 under CVSS 3.1, with high confidentiality impact but only low integrity and availability impact. Versions 22.7R2.14 and earlier are listed as unaffected [2]. Tenable carries the same score and description [3].
The score sits awkwardly beside the outcome. CISA’s ADP entry in the same record marks exploitation as active, automatable and of total technical impact, and records the KEV addition on 13 October [2]. Security Affairs reports a federal remediation deadline of 16 October, three days after listing [6]. Ivanti told reporters it knew of exploitation affecting a limited number of customers [4][5].
The fixed versions differ between sources. The CVE record lists 25.1.3.0, 25.1.3.1 and 22.8R2.5 as unaffected [2]. SecurityWeek also lists 25.1.1.2 as patched [5]. BleepingComputer gives the affected ranges as 22.8R2.4 and earlier, 25.1.2.2 and earlier, and 25.1.1.1 and earlier [4]. Ivanti’s advisory could not be retrieved for this article, so operators should confirm their target build against it directly. Volexity notes that 22.8R2.4 was the latest release when the chain was used against it [1].
What Volexity observed
Volexity began investigating suspicious activity at a network security monitoring customer on 5 October and traced it to the ICS appliance. It dates repeated exploitation by the actor, which it tracks as UTA0576, from 3 October [1]. BleepingComputer reports that Defused saw first signs of exploitation from 1 October [4]. The two start dates differ, and this article could not check the Defused data directly.
Volexity describes the post-exploitation activity as credential-focused preparation for lateral movement:
- Extraction of LDAP domain credentials stored on the appliance [1].
- Active Directory queries for LAPS local administrator passwords, BitLocker recovery keys and attributes that hold passwords [1][5].
- Kerberoasting, credential testing and password spraying against domain accounts [1].
- Packet capture on the appliance to harvest plaintext LDAP and HTTP Basic credentials and NTLM hashes, using a renamed capture tool presented as a system process [1].
- Selective tampering with the runtime debug log and the Linux audit log, self-deleting payloads, and restoring in-memory patches and configuration entries after use [1].
Volexity assesses with high confidence that this was staging for wider network compromise, but says the operation was disrupted before the actor’s objectives became clear [1][5]. It assesses with low confidence that UTA0576 is criminally rather than espionage motivated. It assesses with medium confidence that the actor used large language models for exploit development and post-exploitation scripting, citing the highly specific scripts and the complexity of the chain [1]. These are Volexity’s assessments, not established attribution.
Why the integrity checker misses it
Ivanti’s Integrity Checker Tool (ICT) compares on-disk files against a known-good state. Volexity states that because the chain patches library code in memory only and restores configuration store data afterwards, the tool does not detect any modifications [1]. It warns that the ICT may not identify device compromise and recommends network and memory forensics instead [1]. Ivanti’s public guidance still includes running the ICT regularly [4].
This is not a new limitation. In February 2024 the UK NCSC warned that the ICT is a point-in-time snapshot that cannot necessarily detect activity if an actor has returned the appliance to a clean state, and that it does not scan for malware or indicators of compromise [8]. In April 2025 Mandiant reported that UNC5221, which it assesses as a suspected China-nexus espionage actor, had tried to modify the ICT itself to evade detection [11]. UTA0576 did not need to tamper with the checker. It avoided leaving any artefact for it to check.
The same pattern removes the usual persistence signal. Mandiant observed in-memory droppers that had to be re-executed after a reboot in 2025 [11]. UTA0576 went further by keeping no resident implant at all and relying on continued access to the unpatched chain [1]. Once the appliance is patched, that access route closes. Any credentials already taken remain valid until rotated.
A recurring product, a recurring pattern
ICS has been a repeated zero-day target. CVE-2025-0282, an unauthenticated stack overflow scored 9.0, was published and added to KEV on the same day in January 2025 [10]. CVE-2025-22457 was first patched in February 2025 as what appeared to be a low-risk denial-of-service bug, then exploited from mid-March after, in Mandiant’s assessment, the actor studied the patch [11][12]. In February 2024 a server-side request forgery flaw was used to bypass Ivanti’s interim mitigation for an earlier chain, and Shadowserver counted around 20,800 exposed ICS devices at the time [9]. BleepingComputer counts 36 Ivanti vulnerabilities flagged by CISA as exploited since November 2021 [4].
Current exposure is less clear. BleepingComputer reports that Shadowserver tracks more than 5,500 exposed ICS instances but cannot say how many are honeypots or already secured [4]. That figure could not be checked against Shadowserver’s own data for this article.
What to do
- Upgrade ICS to a build Ivanti’s advisory lists as fixed for CVE-2026-104935, after confirming the version, because published lists differ [2][4][5].
- Treat the appliance as possibly compromised if it was exposed from 1 October onwards. A clean ICT result is not evidence of a clean device [1][8].
- Rotate every credential the appliance stores or brokers, including LDAP bind accounts. Review LAPS and BitLocker key read access, and look for Kerberoasting and spraying from the appliance’s address [1].
- Hunt for lateral movement that starts at the VPN appliance, and keep appliance logs off-box, since local logs were edited [1].
- Watch for high request volumes to the public welcome endpoint and for mismatches between requested paths and internal routing, as Volexity recommends. Use its published indicators [1].
- Plan for the unpatched privilege escalation flaw. Limit management exposure, keep monitoring heightened, and apply the fix when Ivanti releases it, which Volexity expects in the coming weeks [1][5].
For US federal agencies, KEV entries now cite BOD 26-04 and separate forensic triage requirements alongside the remediation deadline [7]. Security Affairs still frames the deadline under BOD 22-01 [6]. Whichever directive applies, the forensic step matters more than usual here, because the patch does not reveal whether the device was already used.
The offensive security reading
The chain shows a practical shift in edge-device tradecraft. The actor kept no implant, so nothing persisted for an integrity check to compare, and it repeated the exploit for each action [1]. That trades stealth on disk for dependence on the vulnerability staying open. It makes patch timing the effective eviction event and makes the credentials taken in the meantime the lasting damage. Red teams assessing VPN estates should test whether defenders can see appliance-originated authentication abuse in the directory, not only whether the appliance shows tampering. Blue teams should assume vendor integrity tooling is one input among several. For an appliance that sits on the identity boundary, memory, network and directory telemetry carry the evidence.
Sources
Every R3KONX article cites its primary material. 12 sources, in order of first citation. Links open the original publication.
- A Series of Unauthenticated Events: Chained Exploitation of Ivanti Connect Secure Zero-Days Volexity · 2026-10-13
- CVE-2026-104935 (CVE record, CNA Ivanti, with CISA ADP container) CVE Program (cvelistV5) · 2026-10-13
- CVE-2026-104935 Tenable · 2026-10-13
- Ivanti warns of Connect Secure zero-day exploited in attacks BleepingComputer · 2026-10-13
- Ivanti zero-day chain gave attackers root access to fully patched VPN SecurityWeek · 2026-10-13
- U.S. CISA adds Ivanti Connect Secure flaw to its Known Exploited Vulnerabilities catalog Security Affairs · 2026-10-13
- Known Exploited Vulnerabilities catalog (JSON feed, catalog version 2026.10.01) CISA via GitHub (cisagov/kev-data) · 2026-10-01
- Exploitation of vulnerabilities affecting Ivanti Connect Secure and Ivanti Policy Secure UK National Cyber Security Centre · 2024-02-09
- Researchers say attackers are mass-exploiting new Ivanti VPN flaw TechCrunch · 2024-02-08
- CVE-2025-0282 (CVE record, CNA Ivanti, with CISA ADP container) CVE Program (cvelistV5) · 2025-01-08
- Suspected China-Nexus Threat Actor Actively Exploiting Critical Ivanti Connect Secure Vulnerability (CVE-2025-22457) Google Threat Intelligence Group / Mandiant · 2025-04-03
- Vulnerability impacting Ivanti Connect Secure, Policy Secure and ZTA Gateways (AL25-004) Canadian Centre for Cyber Security · 2025-04-04
Researched and written by the R3KONX analysis desk from the cited primary material. Ivanti's advisory, CISA's alert page and Shadowserver's data could not be retrieved directly; affected and fixed versions, the federal deadline and exposure counts are taken from the CVE record and the cited reporting. Corrections to event@r3konx.asia.
