Autonomous AI Agent Conducts Breach of Dutch Cybersecurity Nonprofit

The Dutch Institute for Vulnerability Disclosure (DIVD) has suffered a cybersecurity breach involving an autonomous AI agent. The nonprofit organisation, which is composed of volunteer security researchers, described the intrusion as “loud and very, very messy.”

The attack appears to have utilised an agentic AI-powered tool to carry out post-exploitation activities after an initial technical vulnerability was exploited. DIVD researchers noted that the agent operated autonomously, making decisions at high speed that often resulted in “sloppy logic” and pattern errors.

During the incident, the AI agent’s behaviour was observed to be somewhat erratic. Researchers reported that the agent performed actions that appeared poorly executed, such as inadvertently interfering with its own adversary-in-the-middle attack during password spraying attempts. The organisation believes the agent may have been poorly trained or configured for such operations, as it left behind significant evidence and even over-explained its decisions within its own comments.

Investigation and Regulatory Notification

DIVD has launched an investigation into the breach and has notified several authorities, including the police, the Autoriteit Persoonsgegevens (the Dutch data protection authority), and the National Cyber Security Center (NCSC).

While the specific technical vulnerability exploited has not been disclosed, DIVD confirmed it was not related to Citrix NetScaler. The exact purpose of the attack and the full extent of the impact on the organisation’s data remain under investigation.

The organisation has committed to providing a more detailed update on 1 October and aims to notify any other potential victims of the same vulnerability as soon as possible.