AI agent used to breach Dutch vulnerability disclosure nonprofit

The Dutch Institute for Vulnerability Disclosure (DIVD), a volunteer-run nonprofit that finds and reports software flaws, was breached in an attack carried out with the help of an automated AI agent, according to a report published by BleepingComputer on 29 September 2026.

DIVD described the intrusion as "loud and very, very messy", according to the report. That phrasing is the clearest detail available about how the attack behaved. It suggests activity that was noisy enough to be noticed, rather than the quiet, patient approach usually associated with skilled human intruders.

An AI agent, in this context, means software that is given a goal and then chooses and runs its own sequence of steps to reach it, rather than following a fixed script written in advance. Attackers using such tools can generate large volumes of activity quickly, because the agent does not need a person to approve each action.

The report does not say when the attack took place, how long it lasted, or how the attacker first got in. It does not say what systems or data were reached, whether any vulnerability reports or researcher information were exposed, or whether the intrusion has been fully contained. It also does not name the agent or the underlying model used, or identify who was behind the attack. Readers should treat all of those points as unknown rather than assume the worst or the best.

The target is notable in itself. DIVD's role is to receive and pass on information about unpatched security flaws, which means the organisation routinely handles details that would be useful to an attacker. Any incident involving a coordinated disclosure body raises questions about the confidentiality of reports in transit, though nothing in the available account confirms that such material was affected.

The case is one of a small but growing number of reports in which defenders say an intrusion was driven by an agent rather than typed by hand. The practical significance for other organisations is less about novelty and more about detection. If agent-led attacks tend to be noisy, as DIVD's description implies, then volume-based alerting and rate limits may catch them, but the same noise can also swamp small security teams and bury genuine signals.

DIVD has not, in the material available, published a full technical write-up of the incident. Until it does, the sensible reading is narrow: a well-known security nonprofit was successfully attacked, an automated AI agent was part of how that happened, and the organisation chose to say so publicly. Everything else about the case remains unreported.

The Agentic Times will update this story if DIVD publishes further detail.