The short version
- The breach. DIVD, a Dutch nonprofit staffed by volunteer security researchers, says an attacker used two zero-days in the open-source Zammad ticketing system to get into its network. The organisation says the intrusion was carried out by an AI agent that chose its own next steps.
- The flaws. CVE-2026-102489 (CVSS 9.4) lets an unauthenticated attacker run code remotely and leak user sessions; CVE-2026-102490 (CVSS 9.4) lets a local user escalate to root. DIVD says that chained together they allowed session hijacking, remote code execution and root access within seconds.
- What the attacker reached. After the initial compromise the attacker moved to other services and read and exfiltrated data from DIVD's systems. DIVD says network segmentation and its incident response stopped the attacker going deeper, and that its investigation continues.
- Who else is exposed. DIVD says Zammad versions 6.3.0 to 6.5.4 are affected, and that versions 7.0.0 to 7.1.3 contain the defect but cannot be exploited in current conditions. It advises upgrading to version 7 or taking the instance offline, and is scanning for vulnerable instances to warn their owners.
Timeline
| Date | Event |
|---|---|
| 21 September | DIVD's network is breached through the Zammad flaw chain; the organisation starts incident response and notifies the relevant Dutch authorities. |
| 24 September | DIVD says in a LinkedIn post that the modus operandi points to an agentic AI-powered attack. |
| 30 September | DIVD identifies and discloses the two Zammad zero-days; it has published an advisory and a verification script for indicators of compromise. |
How the intrusion worked
DIVD is a nonprofit collective of volunteer security researchers in the Netherlands. It says its network was breached on 21 September and that the entry point was a chain of two previously unknown vulnerabilities in Zammad, an open-source helpdesk and support ticketing platform. The organisation began incident response straight away, blocked access to its infrastructure and notified the relevant Dutch authorities.
The two flaws were identified during the investigation and disclosed on 30 September. CVE-2026-102489, rated 9.4, allows an unauthenticated attacker to execute code remotely and leak user sessions. CVE-2026-102490, also rated 9.4, allows a local user to raise privileges to root. DIVD says the chain let the attacker hijack sessions, run code and escalate from a Zammad user to root in seconds.
From the Zammad instance the attacker reached other services, read data and exfiltrated it. DIVD says segmentation kept the attacker from moving deeper, but that it is still investigating what it found. The flaws were found in collaboration with Merlon Security; DIVD reported them to Zammad, which it says has been working on a fix.
The claim of an AI-driven attack
DIVD's account of the attack rests on how it behaved, not on who was behind it. It has described the intrusion as driven by an AI agent that moved autonomously and picked its next steps without outside direction, and as being, in its words, "loud and very, very messy". In a LinkedIn post on 24 September the organisation said the modus operandi indicated an agentic AI-powered attack.
Part of the evidence came from the attack itself: DIVD says the agent left clear records of its decisions, which let the organisation reconstruct what happened. These are DIVD's own findings about its own network; no source in this material reports an independent assessment of the AI claim.
Who runs a vulnerable instance
Zammad is offered both self-hosted and as a hosted service, and its website claims more than 2,000 customers and 55,000 users. DIVD says versions 6.3.0 to 6.5.4 are affected. Versions 7.0.0 to 7.1.3 also carry the defect, but DIVD says exploitation is not possible there because of the environment conditions.
DIVD's advice is to upgrade to version 7, which it considers safe, or to take the instance offline as soon as possible. It has published a script organisations can run to check for indicators of compromise, and says it is actively scanning for vulnerable Zammad installations and alerting their owners. No source says that any Zammad customer other than DIVD was compromised.
What is not known yet
- Who carried out the attack, and whether a person directed the AI agent or set it running and left it alone.
- How much data was read and exfiltrated from DIVD's systems, and what kind of data it was.
- Whether any other organisation running Zammad was compromised through the same flaw chain.
- When Zammad will release a fixed version covering the affected branches.
- Which other DIVD services the attacker reached before being stopped.
What to do
If you run Zammad
- Check your installed version against 6.3.0 to 6.5.4 and treat any instance in that range as exposed.
- Upgrade to version 7, which DIVD considers safe, or take the instance offline until you can; see Zammad's advisory for the release details.
- Run DIVD's published verification script to look for indicators of compromise on the host.
- Watch DIVD's and Zammad's advisories for further updates before declaring the instance clean.
If you run other internet-facing support or ticketing tools
- Map what the ticketing server can reach and cut network paths it does not need, since segmentation is what stopped this attacker going deeper.
- Log and review authentication events on the ticketing system so that hijacked sessions and privilege escalation to root are visible.
- Keep an inventory of internet-facing services and their versions so a newly disclosed flaw can be matched to your estate quickly.
For everyone else
- Assume a compromise has occurred while an investigation is open rather than waiting for proof, as DIVD did, and keep hunting for signs of compromise until the scope is established.
Update log
- First published.
Sources
New incident reports by email
One email a week with the reports published since the last one.