Four severities, two version ranges, one unfixed half: the Zammad chain
Two Zammad flaws reached CISA's exploited-vulnerability catalogue on 2 October with a 5 October federal deadline. The finder and the vendor do not agree on what is affected or how severe it is, and the privilege-escalation half of the chain had no named fixed version in the vendor's own advisory of 5 October.

The Dutch Institute for Vulnerability Disclosure was breached through its own helpdesk on 21 September 2026 and has published the case in detail [2]. The two vulnerabilities that made it possible are now in CISA’s exploited-vulnerability catalogue [6][7]. The remediation decision is harder than that sounds: the published severity ratings differ by more than a point of CVSS, the affected version ranges differ between finder and vendor, and one half of the chain had no fixed version named in the vendor’s advisory.
The chain
CVE-2026-102489 is a session-handling flaw in Zammad, an open-source helpdesk and ticketing platform. The CVE record describes a session hijack leading to code execution as the zammad service account [6]. CVE-2026-102490 is a local privilege escalation from that account to root [7]. Chained, they take an unauthenticated request to root, which the finder says happened within seconds in its own environment [2].
The timeline is short. First access on 21 September 2026. Detection on 22 September, when the institute blocked all access to its datacentre systems and stood up an incident response team with Merlon Security. Report to the vendor on 24 September, alongside notification of the Dutch data protection authority and the national cyber security centre. On 26 September the institute opened a second case to scan for exposed instances and notify owners. The CVE records were published on 29 and 30 September. CISA added both on 2 October with a due date of 5 October [1][2][6][7]. The vendor’s advisory followed on 5 October [4].
The numbers do not agree
For CVE-2026-102489 the finder, acting as CVE numbering authority, published a CVSS 4.0 score of 8.7, rated high, for the flaw on its own; NVD published a CVSS 3.1 score of 9.8, rated critical; and the finder published a 9.4, rated critical, for the flaw chained with CVE-2026-102490 [6]. For CVE-2026-102490 the finder published 8.5, rated high, against a local-only vector, while NVD published 9.8, critical, against a network vector [7]. One secondary account reported both as 9.4 [10]. Exploit prediction scores sit near one per cent for both [6][7].
The divergence is not cosmetic. NVD’s 9.8 for CVE-2026-102490 uses a network attack vector for a flaw the finder describes as requiring local access and the vendor describes as not remotely exploitable on its own [3][7]. The 9.4 is a chained score, not either flaw’s standalone severity, though it is the figure that drove the critical labelling in coverage [10][11][12]. Anyone using a severity threshold to schedule this work gets a different answer depending on which record their tooling reads.
Which versions, and whose answer
For CVE-2026-102489 the finder lists 6.3.0 through 6.5.4 as affected, and says the vulnerable code is also present in 7.0.0 through 7.1.3 but is not exploitable there because of environmental conditions [1]. The vendor states that the flaw affects 6.5 and earlier, that exploitation depends on the runtime environment those releases use, that 7.0 and later are not affected in practice, and that it hardened the code in 7.2.0 anyway [3]. The CVE record’s description names 7.0.0 to 7.1.2, which does not match the 7.1.3 boundary in its own structured data, and third-party enrichment marks 7.0.0 and later unaffected [6]. One secondary account reported 7.0.0 through 7.1.3 as affected without the qualifier [10]. The practical reading: exposure is concentrated in 6.5 and earlier, which no longer receive security updates at all [3].
For CVE-2026-102490 the range is 1.5.0 to 7.1.0-alpha, which the finder summarises as every version of Zammad including the latest alpha [1][7]. The vendor advisory of 5 October assesses it as a local issue that cannot be exploited remotely on its own, ties it to a confirmed vulnerability in the packaging service, and says it is working on a fix without naming a fixed version [3]. The finder’s case page compounds this by listing patch status as available while its own narrative says the vendor is working on a fix [1].
Where the finder and the vendor part company
On 1 October the vendor published a community statement saying that third-party sites were describing the issue as actively exploited, that the finder had not shared technical details, and that it could not confirm the vulnerability, its scope or the affected versions [5]. The vendor separately states that it received a report on CVE-2026-102489 in August 2026, before the breach, and that information on CVE-2026-102490 was published before it had the detail needed to reproduce it [3]. Treat the vendor advisory as authoritative on fixed versions and the finder’s case pages as authoritative on observed exploitation.
The intruder, as the finder describes it
The institute assesses the intrusion as an agentic, AI-powered attack, on the basis of a modus operandi it says it had not seen before and redacted log screenshots showing attacker scripts carrying notes in which the agent justified its own actions. It describes the activity as loud and messy, with steps chosen automatically on flawed logic, and says it cannot share more without affecting the investigation. It gives no formal confidence level, sees no link to any known threat actor, and says no facts so far support a directly targeted attack on its most sensitive holdings while declining to rule that out [2].
One vendor analysis reads the same evidence as non-deterministic decisions, explanatory comments in scripts, a noisy path, and password spraying that disrupted the attacker’s own interception attempt, and offers poor configuration of the agent as a possibility rather than a finding [9]. No human operator has been identified and no group has claimed the intrusion [9][11]. That is where attribution sits: the finder’s assessment, supported by one third-party reading of partly redacted evidence, not independently verified.
What was taken
As at 1 October the institute confirmed exfiltration of volunteer email addresses and possibly other contact details, without yet determining whose. The helpdesk held correspondence with organisations and researchers seeking help, which the institute warns may include IP addresses of vulnerable systems, vulnerability reports, and portions of credential dumps with masked passwords. Reviews of its collaboration platforms, code hosting and workspace accounts remained open [2][12].
What to do
- Upgrade to 7.2.0 or a later supported release. If you run 6.5 or earlier, treat the instance as unsupported and move now or take it offline [3][5].
- Until a fix for CVE-2026-102490 is published, restrict access to the underlying server to trusted administrators and monitor the service account’s behaviour [3][9].
- Preserve the application and web server logs before rebuilding and run the finder’s indicator-check script against them. A clean result does not rule out compromise [1][8].
- Treat any indication of exploitation as full compromise and rotate every credential held on the host or reachable from it [9].
- Place the helpdesk in its own segment with default-deny egress, which turns exfiltration into a blocked event rather than a detected one [9].
- Alert on behaviour rather than the CVE: the service account spawning interactive shells or connecting out to new hosts, root-owned processes in the application’s process tree, bulk reads of configuration and credential files, and large uploads from the helpdesk segment [9].
The catalogue deadline of 5 October has passed for a chain whose second half has no published patch. That is not a contradiction in the catalogue’s own terms, since the required action is to apply vendor mitigations and the mitigation here is access restriction rather than a version bump [7]. It is the shape of the problem. A helpdesk is a credential concentrator: it accumulates the material an intruder would otherwise have to go looking for, and here it held third parties’ vulnerability reports and partial credential dumps. Classify it at the sensitivity of what passes through it, not that of a ticketing tool. The vendor’s site reports more than 2,000 customers and 55,000 users; how many run an affected release is not published [9].
Sources
Every R3KONX article cites its primary material. 12 sources, in order of first citation. Links open the original publication.
- DIVD-2026-00015: Vulnerabilities in Zammad during investigation of case DIVD-2026-00014 DIVD CSIRT · 2026-10-01
- DIVD-2026-00014: When, not if — DIVD got hacked through AI agents DIVD CSIRT · 2026-10-01
- Security advisory: CVE-2026-102489 and CVE-2026-102490 Zammad · 2026-10-05
- Zammad Security Update on DIVD Case DIVD-2026-00015 (CVE-2026-102489 / CVE-2026-102490) Zammad Community · 2026-10-05
- Zammad statement on reports of active exploitation Zammad Community · 2026-10-01
- CVE-2026-102489 record: scores, affected products and references OpenCVE · 2026-10-07
- CVE-2026-102490 record: scores, affected products and references OpenCVE · 2026-10-07
- CVE-2026-102489: Zammad WebSocket Session Disclosure Horizon3.ai Attack Research · 2026-10-01
- AI agent exploits Zammad zero-days in DIVD breach: what we know and how to detect it Sysdig · 2026-10-06
- U.S. CISA adds Zammad GmbH Zammad flaws to its Known Exploited Vulnerabilities catalog Security Affairs · 2026-10-02
- AI agent exploits zero-day flaws in Zammad ticketing system SC World · 2026-10-01
- Dutch Institute for Vulnerability Disclosure breached via Zammad 0-days Hackread · 2026-10-03
Researched and written by the R3KONX analysis desk from the cited primary material. Three caveats apply. The CISA Known Exploited Vulnerabilities catalogue could not be retrieved directly from this desk, so its entries and dates are reported as the CVE records and secondary accounts describe them. Patch status for CVE-2026-102490 is reported as at the vendor advisory of 5 October 2026 and should be re-checked against the vendor's advisories before acting. The characterisation of the intruder is the finder's own assessment and has not been independently verified. Corrections to event@r3konx.asia.
