The Dutch Institute for Vulnerability Disclosure (DIVD) said the attacker who breached its network got in through two zero-day vulnerabilities in Zammad, the open-source helpdesk and ticketing system it uses, BleepingComputer reported Wednesday. DIVD said the flaws, chained by an autonomous AI agent, took the intruder from a web session to root privileges in seconds.

What Happened

The two vulnerabilities are tracked as CVE-2026-102489 and CVE-2026-102490. Used together, they let the attacker hijack a session, run code remotely and escalate from the account Zammad runs under to root on the server. DIVD attributed the speed of the chain to the agentic nature of the attack: the agent picked its next steps itself, without a human directing each move.

From the Zammad server, the attacker reached other DIVD services and read and exfiltrated data, the nonprofit said. Network segmentation and a quick incident response kept the intrusion from going deeper. DIVD also said the agent left explanations of its choices behind, which helped investigators reconstruct what happened.

RelatedFrance Widens Fridge-Hack Alert to More Controllers as Store Count Passes 350

DIVD urged Zammad users to upgrade to version 7 right away or take vulnerable instances offline until they can. It said the investigation is continuing and promised more detail Thursday. BleepingComputer's report did not include a statement from Zammad.

Why It Matters

The disclosure fills in the biggest gap from DIVD's first account. On Tuesday, as The Primary Wire reported, the group said an autonomous AI agent carried out the post-exploitation phase of the breach, its first security incident in seven years, but it did not name the system that was exploited. Naming Zammad turns a case study about AI-assisted intrusion into an urgent patching issue for other organizations.

Helpdesk platforms are an attractive foothold. They are often exposed to the internet so customers can file tickets, and they hold support conversations, attachments and account details. According to BleepingComputer, Zammad says it has more than 2,000 customers and 55,000 users, including De'Longhi, Amnesty International and Nextcloud, alongside an unknown number of self-hosted open-source installations.

The case also gives defenders an unusually concrete look at how quickly an agent can work once it has a usable exploit chain. The full chain, from hijacked session to root, finished in seconds, leaving little room for alerts to prompt a human response before the attacker moved on. DIVD's experience suggests that basic controls such as segmentation still limit how far such an agent can get.

Context

DIVD is a largely volunteer Dutch nonprofit that scans for exposed and vulnerable systems and warns their owners, which made it a notable target. In its earlier account it said the agent was fast but clumsy, at one point undercutting its own man-in-the-middle attack with password spraying.

What to Watch

DIVD's promised update should say more about what data was taken and how the attacker obtained or built the exploit chain. Zammad administrators should confirm they are on version 7, and researchers will be watching for signs that the same chain is being used against other Zammad installations now that the CVEs are public.