DIVD hacked via two Zammad zero-days

You must now treat patching, segmentation, privilege restriction and logging as a single operational chain rather than separate projects, because automated attackers can combine multiple vulnerabilities faster than human incident response can react.

The prompt is an analysis of 7 October 2026 of agentic-AI attack methods in vulnerability management, which argues that autonomous software can exploit a sequence of flaws in rapid succession, collapsing the time available for human intervention. The concrete case is the compromise of the Dutch Institute for Vulnerability Disclosure on 21 September 2026 via two previously unknown vulnerabilities in Zammad, a helpdesk system the organisation used internally. In our assessment, this incident demonstrates that vulnerability management focused only on identifying flaws is insufficient; the real risk lies in how quickly an autonomous agent can chain multiple weaknesses into a complete compromise—session hijacking, code execution, privilege escalation and data exfiltration—before your team can respond.

How did the attackers move from one vulnerability to the next so quickly?

The two Zammad zero-days formed a deliberate chain. CVE-2026-102489 enabled session hijacking, allowing an attacker to take over an existing session and execute code as the zammad user. CVE-2026-102490 then enabled privilege escalation from that local account to root, granting full system rights. According to the case report, the method matched what is termed an agentic-AI attack—software that autonomously carries out several follow-up steps without waiting for human operator input between each one. This matters because the window between initial compromise and full system control, which might normally span hours or days in a manual attack, can compress to minutes when each step is automated.

Which failure modes does this attack pattern expose?

• Privilege creep in service accounts — a compromised application account retains rights far beyond what that service requires to function, turning a single vulnerability into a path to full system control.
• Absence of segmentation between critical systems — lateral movement from a compromised service to sensitive infrastructure proceeds without friction or detection.
• Slow patch cycles for internet-facing software — publicly reachable systems remain vulnerable for weeks whilst patches are tested and staged, giving automated attackers a wide window.
• Insufficient logging and audit trails — you cannot reconstruct what data an attacker accessed or what actions they performed after compromise.
• Disconnected incident response — security, operations and legal teams do not coordinate on whether to isolate a system immediately or attempt to preserve it for investigation.

What concrete controls must you demonstrate now?

1. Document the rights each service account holds — for every internet-facing application, record which system privileges a compromised account would gain and reduce that account to the minimum required to function.
2. Implement network segmentation for critical systems — ensure that a breach of a public-facing service cannot automatically grant access to internal infrastructure, and test that segmentation regularly.
3. Establish an accelerated patch-or-isolate route — for publicly reachable systems, define timelines based on exposure, exploitability and business impact rather than relying on a standard weeks-long cycle, and ensure taking a system offline is a rehearsed option.
4. Retain independent audit logging — capture execution and access events in a format and location that survives compromise of the primary system, so you can reconstruct which data were accessed.
5. Involve legal and incident teams early — when a publicly reachable system is compromised, assess immediately which personal data may have been accessed and what notification and documentation obligations follow.

Why does treating these controls separately leave you exposed?

Vulnerability management that only determines whether software contains a flaw does not account for how an autonomous agent uses the rights available to a compromised account and chains multiple flaws into a complete attack. The coherence is the control. If your service account holds excessive privileges, segmentation becomes irrelevant because the attacker already has the access they need. If you patch slowly, an automated attacker can exploit the gap. If you do not log independently, you cannot reconstruct what happened or support any legal obligation to notify affected individuals. The DIVD case shows that segmentation and incident response helped limit further lateral movement—that is, the controls worked—but only because they were in place before the breach. Treating them as separate projects means each one is incomplete.

Automated tools can help you identify which privileges each account holds, which systems are segmented and which audit logs are being retained. What tools cannot do is decide which privileges are truly necessary for your business, which systems are critical enough to segment, or which data access patterns matter most to your organisation. That remains your professional judgement, informed by your own risk appetite and operational constraints.