The Dutch Institute for Vulnerability Disclosure (DIVD), a nonprofit that warns organizations about exposed systems, was breached by an attacker who handed the intrusion over to an autonomous AI agent, BleepingComputer reported Tuesday. DIVD said the agent operated on its own inside the network after initial access, choosing its next steps quickly but clumsily.

What Happened

DIVD said it was its first security incident in seven years of operation. The attacker got in by exploiting a vulnerability in a system the organization has not yet named, though DIVD said it was not a Citrix NetScaler flaw. Once inside, the attacker deployed an AI agent that carried out post-exploitation work independently across DIVD's network.

The group's account paints the agent as fast but careless. Its password-spraying attempts interfered with a man-in-the-middle attack it was running at the same time, effectively undercutting its own technique. It also over-explained its decisions in comments on the systems it touched, leaving investigators a detailed record of its reasoning. DIVD said the agent did some plainly foolish things, which suggested it had been poorly trained or configured.

RelatedJPMorgan Bars Outside AI Agents From Customer Accounts as Dimon Warns of 'Rogue Agents'

DIVD called the intrusion "loud and very, very messy," and said the agent decided each next step itself with no apparent human pacing.

The organization has reported the incident to Dutch police, the national data protection authority (Autoriteit Persoonsgegevens) and the National Cyber Security Centre. It is withholding technical details for now to avoid compromising the investigation.

Why It Matters

Security vendors and AI developers have warned for some time that attackers are experimenting with agentic AI to speed up intrusions. Public accounts from victims describing an agent at work inside their own network remain rare, which makes DIVD's report a useful data point on how such attacks behave in practice.

The details cut both ways. An agent that moves at machine speed and picks its own actions can cover ground faster than a human operator. But this one generated noise that defenders could see and left explanations of its own intent, the opposite of the stealth that experienced intrusion teams prize. For defenders, that suggests current agent-driven attacks may be detectable through volume and anomalies, even if they are faster.

The target also matters. DIVD's volunteers scan the internet for vulnerable systems and notify their owners, so the organization holds information about exposed infrastructure and disclosure work in progress. BleepingComputer's report did not say what data, if any, the attacker accessed.

Warning to Other Victims

DIVD said it intends to alert other organizations that may be exposed to the same vulnerability used for initial access. Until the flaw is named, other operators have no way to check whether they are affected, which is one reason the planned disclosure is significant beyond DIVD itself.

What to Watch

DIVD has promised a fuller update on Oct. 1. That report should identify the vulnerability, clarify what the attacker was after and whether data was taken, and may say more about how investigators confirmed an AI agent was involved. Whether the Dutch authorities attribute the attack to a particular actor will also shape how the case is read.