Zammad: How Two Chained Zero-Days (CVE-2026-102489 and 102490) Let an AI Agent Reach Root at DIVD

By Carlos Montiel | Enterprise AI Specialist
Leer en español →
Published: 2026-10-26 | By: Carlos Montiel | Reading time: ~5 minutes
Zammad: How Two Chained Zero-Days (CVE-2026-102489 and 102490) Let an AI Agent Reach Root at DIVD

The Dutch Institute for Vulnerability Disclosure (DIVD), an organization whose job is to warn others about their flaws, was breached through two chained Zammad zero-days. Its own report attributes the speed of the attack to an AI agent that decided its steps without human intervention.

What Happened

According to BleepingComputer and Security Affairs (September 30 and October 1, 2026), DIVD disclosed that its network was compromised through two unpatched vulnerabilities in Zammad, the open-source ticketing and helpdesk platform. DIVD and Merlon Security identified the flaws as CVE-2026-102489 and CVE-2026-102490. Each was serious on its own; chained together, they allowed session hijacking, code execution and escalation from a Zammad user account to root.

From there, the attacker reached other services, read data and exfiltrated some of it. DIVD says it all happened in seconds, and that network segmentation and its incident team's response prevented deeper movement; the investigation was still open. The Register reported that the exposed data included emails from DIVD volunteers.

Why They're Talking About an AI Agent

The attribution rests on behavior: the attack was, in DIVD's words, "noisy and very messy," and the agent left explanations of its decisions that made it possible to reconstruct the incident in detail. That same clumsiness helped detect it.

What isn't confirmed: the Cloud Security Alliance's research note describes the "AI agent" conclusion as an inference from behavior, not a fact verified with the tool in hand. The sources also differ on the CVSS scores (9.4 according to some, 8.7 according to another tracker for CVE-2026-102489), on the affected version ranges, and on whether the starting point was a normal account or unauthenticated access. It should be read as a credible case, not a closed verdict.

How a Zero-Day Chain Works

1. Session: a flaw allows session hijacking in Zammad 2. Execution: from there, remote code execution on the server 3. Escalation: from Zammad user to root on the host 4. Movement: access to other services and reading of data → with an agent, each step is decided and executed in seconds

What's new isn't the technique but the reaction time. A human team takes hours to try, fail and retry; an agent tries variants without pause. The window between "the attacker got in" and "they already have root" shrinks in a way that makes a runbook designed for minutes useless.

What to Do

What this means for your company: if you use Zammad, your patch is the immediate priority. If you don't, the lesson is about design: segmentation, detection in seconds and least privilege matter more when the attacker is automated. DIVD survived the incident because it had segmented; ask your team whether your helpdesk, CRM or wiki are in that position.
Carlos Montiel
Enterprise AI Solutions Architect
Specialist in LLMs, Agents, and Orchestration
guatemalia.com/#contacto · info@guatemalia.com

Need to implement AI in your company?

Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.

Contact Carlos Montiel

info@guatemalia.com