REG-2026-09-24 · 24 SEPTEMBER 2026 Verified

DIVD helpdesk breach — two unknown flaws, chained without a human.

An agent chained two previously unknown flaws in helpdesk software and reached root in seconds, deciding each step itself. Segmentation is what contained it — a control that was already in place before anyone noticed.

Primary source
Disclosing organization · reviewed 24 September 2026
DIVD CSIRT — case DIVD-2026-00015 and accompanying disclosure post
csirt.divd.nl/cases/DIVD-2026-00015

Recorded from the disclosing organization's own published account. DIVD reported the incident to the Dutch data protection authority, the national cyber security centre and the police. Vulnerability identifiers and patch status below follow security reporting of that disclosure dated 1 October 2026 and updated 8 October, and are to be reconciled against the case page at review.

What happened

It decided the next step after every action.

On 21 September 2026 the Dutch Institute for Vulnerability Disclosure — an organization whose work is coordinating the disclosure of vulnerabilities — was itself breached through its helpdesk software. DIVD disclosed the incident on 24 September and later confirmed that two previously unknown flaws in Zammad had been exploited.

The first permitted remote code execution without authentication. The second escalated from the local service account to root. Chained, they took the attacker from no access to root, in DIVD's description, in seconds. From there the attacker reached other services, and read and exfiltrated data.

Automated

DIVD's own account states: "We could see the agent working automated, because after every action it decided the next step itself."

Contained

Network segmentation limited the scope of what the attacker could reach. DIVD's teams prevented deeper movement after detection, while stating that some of the damage was already done.

Unresolved

Neither flaw had a fix at the time of the original disclosure. A mitigation for the first followed in early October; a fix for the second was still in progress.

DIVD states it is assuming breach until it can establish otherwise, and has not said which data was accessed. Whether the intrusion formed part of a wider campaign is not established.

This entry records what the disclosing organization observed about the attacker's behaviour. It does not characterize who directed it, and nothing in the disclosure establishes that.

Which control failed

Every control that needed a person inside the window failed, because there was no window. Patching cannot apply to a flaw nobody has published. Alert triage, escalation and containment all assume a gap between first access and consequence that can be measured in hours. Here the gap between an unauthenticated foothold and root was measured in seconds, and the agent selected each subsequent action itself rather than waiting for an operator.

The second failure is about composition. Neither flaw alone was catastrophic — one gave code execution, one gave escalation from a low-privileged local account. The risk lived in the chain, and the chain was assembled without human direction between the links. Vulnerability management that scores findings individually does not describe that risk at all.

The part worth carrying is what held. Segmentation contained this, and segmentation does not depend on anybody noticing. It was decided before the incident, it was in force during it, and it worked at a speed no responder could have matched. Controls that have to be invoked degrade as attacks accelerate. Controls that are structural do not. That distinction is becoming the whole of the design question.

Zevana exposure analogue

Exposure floors are recorded, not averaged down.

Any organization running internet-facing operational software — a helpdesk, a ticketing system, a portal — carries this structure. The analogue is heightened not because the software was unusual but because it was ordinary: a supporting system, not a crown-jewel application, reached by a route nobody could have patched, at a speed that removed response from the equation. The organization breached here is one that specialises in vulnerability disclosure.

Practice addressed
Strata

Risk & control readiness

Segmentation and blast-radius design, privilege boundaries on service accounts, control mapping that accounts for chained rather than individual findings, and incident readiness measured against the time an automated attack actually takes.

What could go wrong — and what proves our safeguards work?
Ask your organization
One question, this week

Which of our controls would still work if the attack finished before anyone read the alert?

Separate the list into controls that act by themselves and controls that wait for a person. The first list is the one that was load-bearing here, and for most organizations it is the shorter of the two.

← All register entries
Begin with visibility

Ask it of your own systems before someone else does.

A short conversation about what your organization is running, who owns it, and what evidence exists. Scope and fee are confirmed after that conversation — never before.