The incident
The Dutch Institute for Vulnerability Disclosure reported first malicious access to its systems on September 21, 2026. Its investigation identified two Zammad vulnerabilities: CVE-2026-102489, enabling session hijacking and remote code execution as the Zammad user in affected versions, and CVE-2026-102490, enabling the local Zammad user to escalate privileges to root. DIVD has publicly described the attack's modus operandi as involving agentic AI; the breach and vulnerabilities are established, while public evidence does not independently establish the precise model or autonomy level.
The authority problem
An autonomous or compromised system can move from an initial capability to a shell, a more privileged identity, root and additional reachable services. Each step can expand what the system is technically capable of doing without expanding what the enterprise authorized it to do.
Privilege escalation can increase capability. It cannot increase authority.
The authority-layer answer
Enterprise authority should remain bound to the original mandate, delegation, resources and exact effects even as technical privilege changes. Material capability escalation should trigger revalidation rather than silently widening the authority envelope.
GitHub implication
A coding agent might begin with repository authority and later reach a GitHub Actions runner, cloud credential, Kubernetes credential or production environment. The new capability must not inherit the original repository mandate. Infrastructure and deployment effects require their own valid authority.
Why this matters
Machine-speed privilege escalation makes after-the-fact human revocation insufficient. Authority must be enforced at the consequential effect even when the underlying credential or operating-system privilege is technically valid.
Sources
NEXT STEP