AI Agents Exploit PaperCut Vulnerabilities, Compromise 395 Organizations Across 48 Countries
What Happened – A Russian‑speaking threat actor leveraged large‑language‑model agents to automate exploitation of two newly disclosed PaperCut NG/MF flaws (an authentication bypass and a remote‑code‑execution bug). The campaign hit 395 organizations in 48 countries, compromising at least 440 PaperCut servers and obtaining domain‑administrator credentials in 12 cases.
Why It Matters for Trust & Control Assurance
- Demonstrates how unpatched authentication and execution flaws can be weaponised at scale, overwhelming manual patch‑management processes.
- Highlights the need for continuous verification that access‑control mechanisms (least‑privilege, multi‑factor) are enforced and auditable after a vulnerability is disclosed.
- Shows the value of real‑time evidence collection (log aggregation, privileged‑access monitoring) to prove due‑diligence during an audit.
Who Is Affected – Primarily education institutions (≈ 204 victims), but also health, government, and private‑sector organizations that run self‑hosted PaperCut installations.
Recommended Actions
- Deploy PaperCut’s September 10 emergency patches immediately on all on‑premise instances.
- Verify that privileged accounts are protected by MFA and that default or overly‑privileged service accounts are removed or constrained.
- Enable centralized logging of PaperCut and Active Directory events; correlate with threat‑intel feeds for AI‑driven exploit patterns.
- Conduct a rapid control‑assurance review of your access‑control policies and capture evidence for audit readiness.
Technical Notes – The attack chain leveraged:
- Vulnerability 1: authentication bypass (CVE‑2025‑XXXX, CVSS 9.8).
- Vulnerability 2: remote‑code execution (CVE‑2025‑YYYY, CVSS 9.3).
- AI agents built on OpenAI Codex and DeepSeek models to generate and execute exploit code at scale.
- Post‑exploitation steps included credential dumping and lateral movement to Windows domains.
Source: DataBreachToday