Active Exploitation
🔥 CyberForge CVE of the Day #068
🚨 CVE-2026-102489 + CVE-2026-102490 — Zammad: session hijack, service-account execution and a separate route to root
DIVD reports a chain from session hijacking to execution as `zammad`, then local escalation to root, in its September breach. Both CVEs entered CISA KEV on 2 October.
Zammad recommends 7.2.0, but a separate fix for the local escalation is not confirmed in the reviewed sources. Upgrade and compromise investigation remain separate jobs.
🗓️ Evidence cutoff: 3 October 2026, 08:27 UTC / 09:27 BST. This is a dated comparison of CVE/CNA, NVD, CISA, DIVD, Zammad and NCSC-NL records.
🎯 The quick hit:
What is happening? CVE-2026-102489 can lead from a hijacked session to code execution as `zammad`. CVE-2026-102490 concerns that local account gaining root. The second issue needs an existing foothold; it is not independently described by Zammad as a remote entry point.
Who needs to check? Prioritise exposed older deployments; the 6.5.4 boundary is disputed. Zammad says 7.0+ avoids the first flaw's exploitable runtime conditions and adds hardening in 7.2.0. The LPE scope remains unresolved.
First action: identify deployment, build, runtime, exposure and owner. Preserve app, proxy and host evidence before a planned upgrade where feasible. Follow the supported move to 7.2.0; assess isolation/offline operation if timely control is impossible.
Follow this trail: session material in errors → suspicious account activity → unexpected execution → privilege change. Corroborate each step; a log hit does not prove the chain.
Escalate when: session exposure aligns with unexpected execution or privilege use. Missing logs during suspected compromise need an IR decision.
Do not overlook: DIVD supplies a log-check script, but it can print cookies and returns exit 0 even when no logs are found. Read its output and coverage; never automate “exit 0 = safe”.
🔑 Key details — keep the two flaws separate:
CVE-2026-102489 — session hijack to RCE
• Weakness: CWE-384, Session Fixation, in CISA/NVD enrichment.
• DIVD's general CVSS v4.0 score: 8.7 High.
• Network access, no prior privileges, passive user interaction in that assessment.
• Outcome: execution as `zammad`; exact route/parameter undisclosed.
CVE-2026-102490 — local privilege escalation
• Weakness: CWE-269, Improper Privilege Management, in CISA enrichment/KEV.
• DIVD's general CVSS v4.0 score: 8.5 High.
• Local access, low privileges, no user interaction in that assessment.
• Outcome: local `zammad` access can become root.
Combined scenario: DIVD supplies a separate 9.4 Critical CVSS v4.0 assessment in both records, explicitly labelled as chaining with the other CVE:
`CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H/E:A/AU:Y/V:C`.
NVD currently shows that DIVD 9.4 vector for both entries and adds its own CVSS v3.1 9.8 assessment for each: `CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H`. Both records are Analyzed.
NVD's v3.1 says no user interaction; DIVD's first-flaw/chain v4.0 says passive. NVD's network/no-privilege vector for the second CVE conflicts with its local prerequisite. Preserve attribution; do not recast it as independently remote root access.
KEV: both added 2 October 2026, due-date fields 5 October 2026, `forensicTriage: Yes`, ransomware use Unknown. Catalogue 2026.10.02 contained 1,733 entries. The deadlines sit within CISA's federal directive framework and are not a universal private-sector legal deadline.
FIRST EPSS, score date 2 October: 102489 = 0.00582 / 0.582%, percentile 45.835%; 102490 = 0.00262 / 0.262%, percentile 16.336%. These are forecasts and relative rankings, not the chance a particular host is compromised. Known exploitation warrants action despite modest model scores.
🧬 How the reported chain crosses the boundaries.
A session represents application identity; a process runs with OS privileges; root is another boundary. Evidence of one stage does not prove the others occurred locally.
DIVD reports the combined sequence and CISA explicitly says the flaws can be chained. Analyse them together while retaining their prerequisites.
The defensive script checks session material in error output; it does not prove that every logged cookie was obtained by an attacker. The exact request, local escalation primitive and complete intermediate sequence remain undisclosed.
Linux and Docker are listed platforms. For Docker, establish the process namespace: container root is not proof of host root or container escape. Review mounts and privileges when evidence warrants it.
DIVD's AI assessment is not an independently verified actor identity or a CVE detection signature.
📦 Versions, fixes and the disagreements that matter.
First flaw, older releases: DIVD/CNA prose and NCSC-NL include 6.3.0–6.5.4. The CNA machine-readable range is 6.3.0 ≤ version < 6.5.4. NVD also uses an exclusive 6.5.4 upper boundary for that range. Do not mark 6.5.4 safe solely because an inventory feed implements that boundary. The vendor urges users on 6.5 and older to move off unsupported versions.
First flaw, 7.x: DIVD describes the issue as present in 7.0.0–7.1.3 but not exploitable because of environmental conditions; its structured record marks 7.0+ unaffected. Zammad says 7.0+ is not affected and places additional hardening in 7.2.0. NVD nevertheless includes 7.0.0–7.1.3 in its affected configuration data. Record that disagreement; do not turn code presence into demonstrated exploitability.
Second flaw: DIVD prose says all versions including the latest alpha, while its structured range is 1.5.0 ≤ version < 7.1.0-alpha. NVD instead includes versions before 7.1.0 plus an explicit alpha entry. The vendor has not published a definitive affected/fixed matrix in the reviewed statement. A range boundary is not sufficient evidence that 7.1.0 or 7.2.0 fixes the LPE.
Latest reviewed vendor update: on 1 October at 19:48 UTC, Zammad said it had received the LPE details and was working on them; the issue requires existing server access. This supersedes its earlier same-day statement that the details had not arrived. It continued to recommend 7.2.0 and its security-advisory channel.
Practical disposition: move to the vendor-recommended supported release using its upgrade instructions, preserve the second-CVE uncertainty, and track a separate vendor resolution. NCSC-NL's 30 September advisory says the second issue was not yet fixed. DIVD's case-level “patch available” label must not become “both CVEs fixed”.
🕰️ Set the investigation window from your evidence.
DIVD dates first access to 21 September, detection and access blocking to 22 September, and its report to Zammad to 24 September. Those are incident-specific dates. Zammad separately says it first received information about the first issue in August. The disclosure-history disagreement need not delay defence.
CVE publication was 30 September; KEV addition was 2 October. These do not define another organisation's exposure window.
Set the hunt window from local installation/exposure history. Include earlier retained evidence where applicable, record control times and identify rotation, recreation and migration gaps.
🧭 SOC investigation trail.
C01 — Establish the instance and its actual exposure
Action / location: reconcile asset inventory with the application build, package/image identity, runtime, reverse proxy and deployed nodes. Identify the business owner, public routes, internal reachability and historical versions.
Signal / correlation: match each instance to the version caveats and exposure dates; current 7.x does not describe last month's system. Track the LPE separately.
Benign causes / limits: stale inventory, rollback images or host packages can misidentify a container's running build. Verify scanner version logic.
Escalation: exposed legacy deployments go to the service/change owner urgently. Suspicious activity goes to SOC/IR as well. Exposure alone is not a finding of compromise.
C02 — Preserve logs and check their coverage
Action / location: retain application/error logs, reverse-proxy access/error records, available authentication history, container logs and existing host/EDR telemetry. Collect through the approved evidence workflow before planned disruptive changes where feasible.
Signal / correlation: record identity, UTC/clock drift, paths, retention, errors and covered dates. Join request IDs only where the schema provides them.
Benign causes / limits: a quiet file may reflect another log destination, rotation or uncollected stdout. DIVD's defaults are `/var/log/zammad` and `/var/log/nginx`; confirm actual paths, especially in containers.
Escalation: gaps across suspected compromise need an IR decision. Contain active harm promptly and document incomplete collection.
C03 — Hunt DIVD's session-in-error lead privately
Action / location: examine authorised copies of relevant logs for DIVD's published pattern: `ERROR -- :` followed on the same line by `"Cookie"=>"` or `={`.
Signal / correlation: preserve event context/provenance and join account/session activity by time. Use controlled case identifiers; never publish live cookies.
Benign causes / limits: application errors can expose session material without proving hostile use. Redaction, alternate formatting, missing files and expired retention can produce misses. This is a scoped lead, not a complete detector for either vulnerability.
Escalation: investigate session exposure and respond appropriately. A hit alone does not establish RCE or root access.
C04 — Correlate application activity with execution
Action / location: correlate app/proxy records with existing process telemetry for unexpected execution as `zammad`. Retain recorded lineage, paths, command metadata and timestamps.
Signal / correlation: join the suspicious request/session interval to processes on that host/container, then unfamiliar files and egress. Prefer stable event/process IDs.
Benign causes / limits: workers, integrations and maintenance create processes/traffic. Journals are not complete process history; `ruby` or a shell name alone is not an IOC.
Escalation: unexplained execution corroborated by session or network evidence warrants incident handling. Preserve evidence and scope access before deleting suspicious artifacts.
C05 — Test the privilege and data-access hypotheses
Action / location: review existing host audit/EDR, authentication, admin changes and filesystem evidence for unauthorised privilege transitions. Establish container/host boundaries.
Signal / correlation: join identity, lineage and effective privilege. Scope integrations, secrets and data with their owners; egress alone does not prove successful theft.
Benign causes / limits: approved maintenance can produce privileged activity. The undisclosed LPE mechanism means there is no justified universal syscall, sudo event or exploit-file signature to insert here. Missing sudo logs cannot exclude another escalation route.
Escalation: corroborated privilege misuse needs urgent IR review. Suspected integrity loss may require trusted recovery; scope data access from evidence.
C06 — Verify recovery without losing the second CVE
Action / location: verify running build/runtime, migration, service health, exposure controls and forwarding on each node. Assign the unresolved LPE action an owner.
Signal / correlation: reconcile deployment and live build evidence. Track upgrade, session response and incident closure separately.
Benign causes / limits: a healthy login page does not prove every worker restarted or that an old container is gone. New audit features cannot recreate evidence from before installation.
Escalation: failed verification or continuing suspicious activity returns to IR. Closure needs findings, coverage and residual uncertainty.
Handover fields: instance/owner → UTC window → source and record IDs → finding/confidence → gaps → action/owner. Distinguish exposed, suspicious, corroborated compromise and insufficient telemetry.
🔎 What the published checker can and cannot tell you.
DIVD links cve-2026-102489_ioc_check_script_v2.sh from its case page. This defensive checker searches plain/compressed app and nginx logs for the pattern above; it is not an exploit.
We reviewed that script and exercised a line-ending-normalised working copy against synthetic match, non-match and missing-file fixtures. It returned 0 in all three cases, including “no log files found”. Its scan also suppresses some read errors; a no-indicator message can therefore conceal coverage problems if you do not verify the inputs.
It can print distinct cookie values, limits the displayed matching lines and truncates displayed line content. Treat its output as sensitive and retain full source records separately. The script's SHA-256 in our evidence register identifies the reviewed tool; it is not a malware hash or IOC to deploy as a detection rule.
No verified campaign IP/domain list or malicious-file hash set was established. Use the published log lead and labelled behavioural correlations.
🧰 Three bounded checks for an authorised investigation.
Check A — Installed package metadata
For a Debian/Ubuntu package installation, with permission to inspect that host's package database:
```bash
dpkg-query -W -f='${Package}\t${Version}\t${Status}\n' zammad
```
Read package/version/status, then verify the running app and runtime. A missing package may mean Docker or another deployment. This does not establish compromise. Syntax was help/manual-checked; no Zammad host was queried.
Check B — Locate candidate log files without printing cookies
Use ripgrep on an approved plain-text working copy. Replace the path and select the UTC window during collection; this search has no date filter. Decompress working copies first.
```bash
rg -l -- 'ERROR -- :.*("Cookie"=>"|=\{)' \
/analysis/zammad-case/logs/production.log
```
The expression follows DIVD's published pattern. `-l` prints matching filenames, not matching lines or cookie contents. Review the original events privately after a hit. Exit 0 means a match, 1 no match, 2 an error; resolve errors and coverage before interpreting a negative. More filenames can be supplied explicitly after the pattern.
This exact regex and filename-only behaviour passed synthetic positive, negative and missing-file checks with ripgrep 15.2.0. These are local syntax/behaviour checks, not validation against a compromised Zammad instance. Different log formats or redaction require careful adaptation.
Check C — Read a bounded Docker Compose log interval
For an authorised existing Zammad Compose deployment, first confirm the correct project/context and its actual service names. The upstream stack defines `zammad-railsserver` and `zammad-scheduler`; custom deployments can differ. Example one-hour interval, to replace with the case window:
```bash
docker compose logs --no-color --timestamps \
--since '2026-09-21T08:00:00Z' \
--until '2026-09-21T09:00:00Z' \
zammad-railsserver zammad-scheduler
```
This reads logs without restarting containers. Docker access must already be authorised; output can be sensitive. The illustrative hour is not a claimed attack time and omits other services.
Recreated containers, logging drivers and retention can limit what remains. Inspect errors and source coverage; normal output is not a guarantee of full history. Flags were checked against Docker documentation and local Compose 2.40.3-3 help; no client stack was queried.
🩹 Remediate, verify and investigate the earlier window.
1 — Assign owners and preserve evidence. Service, SOC/IR and platform owners need one shared timeline. NCSC-NL explicitly recommends retaining application and network logs before updating. Preserve relevant host evidence too where available; active harm may require immediate containment.
2 — Follow the supported route to 7.2.0. Plan dependencies, backup/restore and migration; do not skip major versions. If timely control is impossible, assess access restriction/offline operation with the owner. These affect availability and do not remove an existing foothold.
3 — Keep CVE-2026-102490 open for vendor resolution. Record the latest statement and the uncertainty about affected/fixed builds. A move to a runtime where the first step is not exploitable reduces that route; it does not prove the local issue was fixed or prevent every other way to obtain local access.
4 — Respond to session or host findings. When session exposure or compromise is supported, use the application's supported revocation process and scope credentials/integrations with IR. Do not assume a password change invalidates every existing session. Avoid publishing sensitive token output or destroying evidence through casual cleanup.
5 — Establish trust. Verify builds, workers, service health and controls. Investigate earlier activity; uncertain host integrity may require IR-led trusted recovery.
Acceptance evidence: running build, service/control checks, retained evidence, explicit LPE disposition and a separate incident conclusion.
🧾 Evidence separation.
Published primary evidence: DIVD's descriptions and incident report, CISA KEV entries, the CVE/NVD score and version fields, Zammad's dated statements and recommended release, and DIVD's defensive script.
Unresolved: boundary builds, LPE scope/fix, exploit mechanics and record disagreements. No working public exploit was validated. Unreviewed mirrors do not resolve these gaps.
CyberForge analysis: follow the session-to-process-to-privilege evidence trail, protect exposed tokens, and close the upgrade and compromise assessment separately. These are investigation recommendations, not findings on a customer's system.
🔥 CyberForge verdict.
This duo deserves urgent attention because exploitation is documented and the consequences cross application and operating-system boundaries. The latest vendor guidance provides a concrete upgrade direction, while the second flaw still needs an explicit resolution.
Aim for a verified supported deployment, an evidence-led account of earlier activity and a named owner for the remaining uncertainty.
> Move the application forward. Trace what happened behind it. Keep the local-escalation question open until the evidence closes it.
🔗 Sources and further reading.
DIVD vulnerability case and linked log checker:
https://csirt.divd.nl/cases/DIVD-2026-00015/
Latest reviewed Zammad update:
https://community.zammad.org/t/21297/3
Zammad 7.2 and supported update instructions:
https://zammad.com/en/product/releases/7-2
https://docs.zammad.org/en/latest/install/update.html
CVE records:
https://www.cve.org/CVERecord?id=CVE-2026-102489
https://www.cve.org/CVERecord?id=CVE-2026-102490
NVD:
https://nvd.nist.gov/vuln/detail/CVE-2026-102489
https://nvd.nist.gov/vuln/detail/CVE-2026-102490
CISA KEV:
https://www.cisa.gov/known-exploited-vulnerabilities-catalog
NCSC-NL advisory:
https://advisories.ncsc.nl/2026/ncsc-2026-0396.html
FIRST EPSS:
https://api.first.org/data/v1/epss?cve=CVE-2026-102489,CVE-2026-102490
Docker log-query reference:
https://docs.docker.com/reference/cli/docker/compose/logs/