Two Zero-Days Exploited in Attack on Dutch Institute for Vulnerability Disclosure

A Dutch cybersecurity non-profit, the Institute for Vulnerability Disclosure (DIVD), was compromised following an agentic AI attack exploiting two zero-day flaws in its Zammad helpdesk platform. The attackers used these vulnerabilities to hijack sessions, execute remote code, and escalate privileges from a Zammad user to root, facilitated by the agentic nature of the exploit. This access allowed the attackers to reach other services and exfiltrate data. The exploited vulnerabilities were remote code execution bug CVE-2026-102489 and elevation of privileges flaw CVE-2026-102490, both scoring 9.4 when chained together. Despite the compromise, DIVD's security team contained the threat through network segmentation and incident response actions, although some damage occurred. A separate data file indicated that volunteer data, including email addresses, was potentially compromised, raising concerns about impersonation risks. The investigation suggested the attack involved an agentic AI, as attacker scripts included justifications for their actions. Experts emphasize that while AI accelerates attacks, fundamental security practices like segmentation, monitoring, and incident response remain critical.

The narrative structure highlights the tension between the speed of modern adversarial capabilities—exemplified by agentic AI attacks—and the necessary, slower requirements for robust security architecture. The initial exploitation leverages a high-speed, automated method (agentic hacking) to bypass traditional human-centric detection, immediately shifting the focus from vulnerability patching (the input vector) to systemic defense (the containment strategy). This creates a pattern where novel vulnerabilities are weaponized instantly, demanding that defensive measures evolve beyond static patch management toward dynamic system segmentation and continuous visibility. The mention of agentic justification within the attacker’s logs points to an emerging paradigm where machine autonomy can introduce a layer of obfuscation that evades standard forensic review; this suggests that accountability shifts from identifying human operators to understanding autonomous decision-making chains in security events. The warning emphasizes that foundational principles—segmentation, monitoring, and response protocols—are not obsolete but are amplified by AI speed, suggesting that organizational resilience depends less on stopping the *next* zero-day exploit and more on establishing unbreakable boundaries where even automated processes cannot traverse freely. What is the implicit cost placed on volunteer organizations when they become targets for sophisticated state-level or advanced criminal actors using these emerging tools? How does relying on human incident response teams, even when effective, adapt to an environment where machine-driven speed compresses the window for meaningful intervention?